Add POI feature and imporve the overall backend implementation

services/external/README.md

@@ -42,12 +42,22 @@ The **External Service** integrates real-world data from Spanish sources to enha
 - **Rate Limit Management** - Respect API usage limits
 - **Error Logging** - Detailed error tracking and alerts
 
+### POI Context Detection
+- **OpenStreetMap Integration** - Automatic detection of Points of Interest (POIs) near bakery locations
+- **18+ POI Categories** - Schools, offices, transport hubs, residential areas, commercial zones, and more
+- **Competitive Intelligence** - Identify nearby competing bakeries and complementary businesses
+- **Location-Based Features** - Generate ML features from POI proximity and density
+- **High-Impact Categories** - Automatically identify POI categories most relevant to bakery demand
+- **Caching & Optimization** - PostgreSQL storage with spatial indexing for fast retrieval
+- **Onboarding Integration** - Automatic POI detection during bakery setup
+
 ### Feature Engineering
 - **Weather Impact Scores** - Calculate weather influence on demand
 - **Traffic Influence** - Quantify traffic effect on foot traffic
 - **Holiday Types** - Categorize holidays by demand impact
 - **Season Detection** - Identify seasonal patterns
 - **Weekend vs. Weekday** - Business day classification
+- **POI Features** - Location context features from nearby points of interest
 - **Combined Features** - Multi-factor feature generation
 
 ## Business Value
@@ -89,6 +99,16 @@ The **External Service** integrates real-world data from Spanish sources to enha
 
 ## API Endpoints (Key Routes)
 
+### POI Detection & Context
+- `POST /poi-context/{tenant_id}/detect` - Detect POIs for tenant location (lat, long, force_refresh params)
+- `GET /poi-context/{tenant_id}` - Get cached POI context for tenant
+- `POST /poi-context/{tenant_id}/refresh` - Force refresh POI detection
+- `DELETE /poi-context/{tenant_id}` - Delete POI context for tenant
+- `GET /poi-context/{tenant_id}/feature-importance` - Get POI feature importance summary
+- `GET /poi-context/{tenant_id}/competitor-analysis` - Get competitive analysis
+- `GET /poi-context/health` - Check POI service and Overpass API health
+- `GET /poi-context/cache/stats` - Get POI cache statistics
+
 ### Weather Data (AEMET)
 - `GET /api/v1/external/weather/current` - Current weather for location
 - `GET /api/v1/external/weather/forecast` - 7-day weather forecast
@@ -121,10 +141,47 @@ The **External Service** integrates real-world data from Spanish sources to enha
 - `GET /api/v1/external/health/traffic` - Traffic API status
 - `GET /api/v1/external/metrics` - API usage metrics
 
+### Geocoding (Address Lookup)
+- `GET /api/v1/geocoding/search` - Search addresses with autocomplete
+- `GET /api/v1/geocoding/reverse` - Reverse geocode coordinates to address
+- `GET /api/v1/geocoding/health` - Check geocoding service health
+
 ## Database Schema
 
 ### Main Tables
 
+**tenant_poi_contexts** (POI Data Storage)
+```sql
+CREATE TABLE tenant_poi_contexts (
+    id UUID PRIMARY KEY,
+    tenant_id UUID NOT NULL UNIQUE,                     -- Tenant reference
+    bakery_location JSONB NOT NULL,                     -- {latitude, longitude, address}
+
+    -- POI detection results
+    poi_detection_results JSONB NOT NULL,               -- Full detection results by category
+    total_pois_detected INTEGER DEFAULT 0,
+    relevant_categories TEXT[],                          -- Categories with POIs nearby
+    high_impact_categories TEXT[],                       -- High-relevance categories
+
+    -- ML features for forecasting
+    ml_features JSONB,                                   -- Pre-computed ML features
+
+    -- Competitive analysis
+    competitive_insights JSONB,                          -- Nearby bakeries and competition
+
+    -- Metadata
+    detected_at TIMESTAMP DEFAULT NOW(),
+    last_refreshed_at TIMESTAMP,
+    detection_radius_meters INTEGER DEFAULT 1000,
+    osm_data_timestamp TIMESTAMP,
+
+    created_at TIMESTAMP DEFAULT NOW(),
+    updated_at TIMESTAMP DEFAULT NOW()
+);
+
+CREATE INDEX idx_tenant_poi_tenant ON tenant_poi_contexts(tenant_id);
+```
+
 **weather_data**
 ```sql
 CREATE TABLE weather_data (
@@ -907,17 +964,21 @@ python main.py
 - **AEMET API** - Spanish weather data
 - **Madrid Open Data** - Traffic data
 - **Spanish Government** - Holiday calendar
+- **OpenStreetMap Overpass API** - POI detection data source
+- **Nominatim API** - Geocoding and address lookup
 - **Auth Service** - User authentication
-- **PostgreSQL** - External data storage
+- **PostgreSQL** - External data storage (weather, traffic, holidays, POIs)
 - **Redis** - API response caching
 - **RabbitMQ** - Event publishing
 
 ### Dependents
-- **Forecasting Service** - Uses external features for ML models
-- **AI Insights Service** - Weather/holiday-based recommendations
+- **Forecasting Service** - Uses external features (weather, traffic, holidays, POI features) for ML models
+- **Training Service** - Fetches POI features during model training
+- **AI Insights Service** - Weather/holiday/location-based recommendations
 - **Production Service** - Weather-aware production planning
 - **Notification Service** - Holiday and weather alerts
-- **Frontend Dashboard** - Display weather and holidays
+- **Frontend Dashboard** - Display weather, holidays, and POI detection results
+- **Onboarding Flow** - Automatic POI detection during bakery setup
 
 ## Business Value for VUE Madrid
 

services/external/app/api/geocoding.py

@@ -0,0 +1,302 @@
+"""
+Geocoding API Endpoints
+
+Provides address search, autocomplete, and geocoding via Nominatim.
+"""
+
+from fastapi import APIRouter, Query, HTTPException
+from typing import List, Optional
+from pydantic import BaseModel, Field
+import structlog
+
+from app.services.nominatim_service import NominatimService
+
+logger = structlog.get_logger()
+router = APIRouter(prefix="/api/v1/geocoding", tags=["Geocoding"])
+
+# Initialize Nominatim service
+# In production, override with environment variable for self-hosted instance
+nominatim_service = NominatimService()
+
+
+# Response Models
+class AddressResult(BaseModel):
+    """Address search result"""
+    display_name: str = Field(..., description="Full formatted address")
+    lat: float = Field(..., description="Latitude")
+    lon: float = Field(..., description="Longitude")
+    osm_type: str = Field(..., description="OSM object type")
+    osm_id: int = Field(..., description="OSM object ID")
+    place_id: int = Field(..., description="Nominatim place ID")
+    type: str = Field(..., description="Place type")
+    class_: str = Field(..., alias="class", description="OSM class")
+    address: dict = Field(..., description="Parsed address components")
+    boundingbox: List[str] = Field(..., description="Bounding box coordinates")
+
+
+class GeocodeResult(BaseModel):
+    """Geocoding result"""
+    display_name: str = Field(..., description="Full formatted address")
+    lat: float = Field(..., description="Latitude")
+    lon: float = Field(..., description="Longitude")
+    address: dict = Field(..., description="Parsed address components")
+
+
+class CoordinateValidation(BaseModel):
+    """Coordinate validation result"""
+    valid: bool = Field(..., description="Whether coordinates are valid")
+    address: Optional[str] = Field(None, description="Address at coordinates if valid")
+
+
+# Endpoints
+@router.get(
+    "/search",
+    response_model=List[AddressResult],
+    summary="Search for addresses",
+    description="Search for addresses matching query (autocomplete). Minimum 3 characters required."
+)
+async def search_addresses(
+    q: str = Query(..., min_length=3, description="Search query (minimum 3 characters)"),
+    country_code: str = Query("es", description="ISO country code to restrict search"),
+    limit: int = Query(10, ge=1, le=50, description="Maximum number of results")
+):
+    """
+    Search for addresses matching the query.
+
+    This endpoint provides autocomplete functionality for address input.
+    Results are restricted to the specified country and sorted by relevance.
+
+    Example:
+        GET /api/v1/geocoding/search?q=Gran%20Via%20Madrid&limit=5
+    """
+    try:
+        results = await nominatim_service.search_address(
+            query=q,
+            country_code=country_code,
+            limit=limit
+        )
+
+        logger.info(
+            "Address search request",
+            query=q,
+            country=country_code,
+            result_count=len(results)
+        )
+
+        return results
+
+    except Exception as e:
+        logger.error(
+            "Address search failed",
+            query=q,
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"Address search failed: {str(e)}"
+        )
+
+
+@router.get(
+    "/geocode",
+    response_model=GeocodeResult,
+    summary="Geocode an address",
+    description="Convert an address string to coordinates (lat/lon)"
+)
+async def geocode_address(
+    address: str = Query(..., min_length=5, description="Full address to geocode"),
+    country_code: str = Query("es", description="ISO country code")
+):
+    """
+    Geocode an address to get coordinates.
+
+    Returns the best matching location for the given address.
+
+    Example:
+        GET /api/v1/geocoding/geocode?address=Gran%20Via%2028,%20Madrid
+    """
+    try:
+        result = await nominatim_service.geocode_address(
+            address=address,
+            country_code=country_code
+        )
+
+        if not result:
+            raise HTTPException(
+                status_code=404,
+                detail=f"Address not found: {address}"
+            )
+
+        logger.info(
+            "Geocoding request",
+            address=address,
+            lat=result["lat"],
+            lon=result["lon"]
+        )
+
+        return result
+
+    except HTTPException:
+        raise
+    except Exception as e:
+        logger.error(
+            "Geocoding failed",
+            address=address,
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"Geocoding failed: {str(e)}"
+        )
+
+
+@router.get(
+    "/reverse",
+    response_model=GeocodeResult,
+    summary="Reverse geocode coordinates",
+    description="Convert coordinates (lat/lon) to an address"
+)
+async def reverse_geocode(
+    lat: float = Query(..., ge=-90, le=90, description="Latitude"),
+    lon: float = Query(..., ge=-180, le=180, description="Longitude")
+):
+    """
+    Reverse geocode coordinates to get address.
+
+    Returns the address at the specified coordinates.
+
+    Example:
+        GET /api/v1/geocoding/reverse?lat=40.4168&lon=-3.7038
+    """
+    try:
+        result = await nominatim_service.reverse_geocode(
+            latitude=lat,
+            longitude=lon
+        )
+
+        if not result:
+            raise HTTPException(
+                status_code=404,
+                detail=f"No address found at coordinates: {lat}, {lon}"
+            )
+
+        logger.info(
+            "Reverse geocoding request",
+            lat=lat,
+            lon=lon,
+            address=result["display_name"]
+        )
+
+        return result
+
+    except HTTPException:
+        raise
+    except Exception as e:
+        logger.error(
+            "Reverse geocoding failed",
+            lat=lat,
+            lon=lon,
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"Reverse geocoding failed: {str(e)}"
+        )
+
+
+@router.get(
+    "/validate",
+    response_model=CoordinateValidation,
+    summary="Validate coordinates",
+    description="Check if coordinates point to a valid location"
+)
+async def validate_coordinates(
+    lat: float = Query(..., ge=-90, le=90, description="Latitude"),
+    lon: float = Query(..., ge=-180, le=180, description="Longitude")
+):
+    """
+    Validate that coordinates point to a real location.
+
+    Returns validation result with address if valid.
+
+    Example:
+        GET /api/v1/geocoding/validate?lat=40.4168&lon=-3.7038
+    """
+    try:
+        is_valid = await nominatim_service.validate_coordinates(
+            latitude=lat,
+            longitude=lon
+        )
+
+        result = {"valid": is_valid, "address": None}
+
+        if is_valid:
+            geocode_result = await nominatim_service.reverse_geocode(lat, lon)
+            if geocode_result:
+                result["address"] = geocode_result["display_name"]
+
+        logger.info(
+            "Coordinate validation request",
+            lat=lat,
+            lon=lon,
+            valid=is_valid
+        )
+
+        return result
+
+    except Exception as e:
+        logger.error(
+            "Coordinate validation failed",
+            lat=lat,
+            lon=lon,
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"Coordinate validation failed: {str(e)}"
+        )
+
+
+@router.get(
+    "/health",
+    summary="Check geocoding service health",
+    description="Check if Nominatim service is accessible"
+)
+async def health_check():
+    """
+    Check if Nominatim service is accessible.
+
+    Returns service health status.
+    """
+    try:
+        is_healthy = await nominatim_service.health_check()
+
+        if not is_healthy:
+            raise HTTPException(
+                status_code=503,
+                detail="Nominatim service is unavailable"
+            )
+
+        return {
+            "status": "healthy",
+            "service": "nominatim",
+            "base_url": nominatim_service.base_url,
+            "is_public_api": nominatim_service.is_public_api
+        }
+
+    except HTTPException:
+        raise
+    except Exception as e:
+        logger.error(
+            "Health check failed",
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=503,
+            detail=f"Health check failed: {str(e)}"
+        )

services/external/app/api/poi_context.py

@@ -0,0 +1,452 @@
+"""
+POI Context API Endpoints
+
+REST API for POI detection, retrieval, and management.
+"""
+
+from fastapi import APIRouter, Depends, HTTPException, Query
+from sqlalchemy.ext.asyncio import AsyncSession
+from typing import Optional
+import structlog
+import uuid
+
+from app.core.database import get_db
+from app.services.poi_detection_service import POIDetectionService
+from app.services.poi_feature_selector import POIFeatureSelector
+from app.services.competitor_analyzer import CompetitorAnalyzer
+from app.services.poi_refresh_service import POIRefreshService
+from app.repositories.poi_context_repository import POIContextRepository
+from app.cache.poi_cache_service import POICacheService
+from app.core.redis_client import get_redis_client
+
+logger = structlog.get_logger()
+
+router = APIRouter(prefix="/poi-context", tags=["POI Context"])
+
+
+@router.post("/{tenant_id}/detect")
+async def detect_pois_for_tenant(
+    tenant_id: str,
+    latitude: float = Query(..., description="Bakery latitude"),
+    longitude: float = Query(..., description="Bakery longitude"),
+    force_refresh: bool = Query(False, description="Force refresh, skip cache"),
+    db: AsyncSession = Depends(get_db)
+):
+    """
+    Detect POIs for a tenant's bakery location.
+
+    Performs automated POI detection using Overpass API, calculates ML features,
+    and stores results for demand forecasting.
+    """
+    try:
+        tenant_uuid = uuid.UUID(tenant_id)
+    except ValueError:
+        raise HTTPException(status_code=400, detail="Invalid tenant_id format")
+
+    logger.info(
+        "POI detection requested",
+        tenant_id=tenant_id,
+        location=(latitude, longitude),
+        force_refresh=force_refresh
+    )
+
+    try:
+        # Initialize services
+        poi_service = POIDetectionService()
+        feature_selector = POIFeatureSelector()
+        competitor_analyzer = CompetitorAnalyzer()
+        poi_repo = POIContextRepository(db)
+        redis_client = await get_redis_client()
+        cache_service = POICacheService(redis_client)
+
+        # Check cache first (unless force refresh)
+        if not force_refresh:
+            cached_result = await cache_service.get_cached_pois(latitude, longitude)
+            if cached_result:
+                logger.info("Using cached POI results", tenant_id=tenant_id)
+                # Still save to database for this tenant
+                poi_context = await poi_repo.create_or_update(tenant_uuid, cached_result)
+                return {
+                    "status": "success",
+                    "source": "cache",
+                    "poi_context": poi_context.to_dict()
+                }
+
+        # Detect POIs
+        poi_results = await poi_service.detect_pois_for_bakery(
+            latitude, longitude, tenant_id
+        )
+
+        # Select relevant features
+        try:
+            feature_selection = feature_selector.select_relevant_features(
+                poi_results["poi_categories"],
+                tenant_id
+            )
+        except Exception as e:
+            logger.error(
+                "Feature selection failed", 
+                tenant_id=tenant_id, 
+                error=str(e),
+                exc_info=True
+            )
+            # Provide default feature selection to continue
+            feature_selection = {
+                "features": {},
+                "relevant_categories": [],
+                "relevance_report": [],
+                "total_features": 0,
+                "total_relevant_categories": 0
+            }
+
+        # Analyze competitors specifically
+        try:
+            competitors_data = poi_results["poi_categories"].get("competitors", {})
+            competitor_pois = competitors_data.get("pois", [])
+            competitor_analysis = competitor_analyzer.analyze_competitive_landscape(
+                competitor_pois,
+                (latitude, longitude),
+                tenant_id
+            )
+        except Exception as e:
+            logger.error(
+                "Competitor analysis failed", 
+                tenant_id=tenant_id, 
+                error=str(e),
+                exc_info=True
+            )
+            # Provide default competitor analysis to continue
+            competitor_analysis = {
+                "competitive_pressure_score": 0.0,
+                "direct_competitors_count": 0,
+                "nearby_competitors_count": 0,
+                "market_competitors_count": 0,
+                "total_competitors_count": 0,
+                "competitive_zone": "low_competition",
+                "market_type": "underserved",
+                "competitive_advantage": "first_mover",
+                "ml_feature_competitive_pressure": 0.0,
+                "ml_feature_has_direct_competitor": 0,
+                "ml_feature_competitor_density_500m": 0,
+                "competitor_details": [],
+                "nearest_competitor": None
+            }
+
+        # Generate competitive insights
+        try:
+            competitive_insights = competitor_analyzer.get_competitive_insights(
+                competitor_analysis
+            )
+        except Exception as e:
+            logger.warning(
+                "Failed to generate competitive insights", 
+                tenant_id=tenant_id, 
+                error=str(e)
+            )
+            competitive_insights = []
+
+        # Combine results
+        enhanced_results = {
+            **poi_results,
+            "ml_features": feature_selection.get("features", {}),
+            "relevant_categories": feature_selection.get("relevant_categories", []),
+            "relevance_report": feature_selection.get("relevance_report", []),
+            "competitor_analysis": competitor_analysis,
+            "competitive_insights": competitive_insights
+        }
+
+        # Cache results
+        try:
+            await cache_service.cache_poi_results(latitude, longitude, enhanced_results)
+        except Exception as e:
+            logger.warning(
+                "Failed to cache POI results", 
+                tenant_id=tenant_id, 
+                error=str(e)
+            )
+
+        # Save to database
+        try:
+            poi_context = await poi_repo.create_or_update(tenant_uuid, enhanced_results)
+        except Exception as e:
+            logger.error(
+                "Failed to save POI context to database", 
+                tenant_id=tenant_id, 
+                error=str(e),
+                exc_info=True
+            )
+            raise HTTPException(
+                status_code=500,
+                detail=f"Failed to save POI context: {str(e)}"
+            )
+
+        # Schedule automatic refresh job (180 days from now)
+        try:
+            poi_refresh_service = POIRefreshService()
+            refresh_job = await poi_refresh_service.schedule_refresh_job(
+                tenant_id=tenant_id,
+                latitude=latitude,
+                longitude=longitude,
+                session=db
+            )
+            logger.info(
+                "POI refresh job scheduled",
+                tenant_id=tenant_id,
+                job_id=str(refresh_job.id),
+                scheduled_at=refresh_job.scheduled_at
+            )
+        except Exception as e:
+            logger.warning(
+                "Failed to schedule POI refresh job",
+                tenant_id=tenant_id,
+                error=str(e)
+            )
+
+        logger.info(
+            "POI detection completed",
+            tenant_id=tenant_id,
+            total_pois=poi_context.total_pois_detected,
+            relevant_categories=len(feature_selection.get("relevant_categories", []))
+        )
+
+        return {
+            "status": "success",
+            "source": "detection",
+            "poi_context": poi_context.to_dict(),
+            "feature_selection": feature_selection,
+            "competitor_analysis": competitor_analysis,
+            "competitive_insights": competitive_insights
+        }
+
+    except Exception as e:
+        logger.error(
+            "POI detection failed",
+            tenant_id=tenant_id,
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"POI detection failed: {str(e)}"
+        )
+
+
+@router.get("/{tenant_id}")
+async def get_poi_context(
+    tenant_id: str,
+    db: AsyncSession = Depends(get_db)
+):
+    """
+    Get POI context for a tenant.
+
+    Returns stored POI detection results and ML features.
+    """
+    try:
+        tenant_uuid = uuid.UUID(tenant_id)
+    except ValueError:
+        raise HTTPException(status_code=400, detail="Invalid tenant_id format")
+
+    poi_repo = POIContextRepository(db)
+    poi_context = await poi_repo.get_by_tenant_id(tenant_uuid)
+
+    if not poi_context:
+        raise HTTPException(
+            status_code=404,
+            detail=f"POI context not found for tenant {tenant_id}"
+        )
+
+    # Check if stale
+    is_stale = poi_context.is_stale()
+
+    return {
+        "poi_context": poi_context.to_dict(),
+        "is_stale": is_stale,
+        "needs_refresh": is_stale
+    }
+
+
+@router.post("/{tenant_id}/refresh")
+async def refresh_poi_context(
+    tenant_id: str,
+    db: AsyncSession = Depends(get_db)
+):
+    """
+    Refresh POI context for a tenant.
+
+    Re-detects POIs and updates stored data.
+    """
+    try:
+        tenant_uuid = uuid.UUID(tenant_id)
+    except ValueError:
+        raise HTTPException(status_code=400, detail="Invalid tenant_id format")
+
+    poi_repo = POIContextRepository(db)
+    existing_context = await poi_repo.get_by_tenant_id(tenant_uuid)
+
+    if not existing_context:
+        raise HTTPException(
+            status_code=404,
+            detail=f"POI context not found for tenant {tenant_id}. Use detect endpoint first."
+        )
+
+    # Perform detection with force_refresh=True
+    return await detect_pois_for_tenant(
+        tenant_id=tenant_id,
+        latitude=existing_context.latitude,
+        longitude=existing_context.longitude,
+        force_refresh=True,
+        db=db
+    )
+
+
+@router.delete("/{tenant_id}")
+async def delete_poi_context(
+    tenant_id: str,
+    db: AsyncSession = Depends(get_db)
+):
+    """
+    Delete POI context for a tenant.
+    """
+    try:
+        tenant_uuid = uuid.UUID(tenant_id)
+    except ValueError:
+        raise HTTPException(status_code=400, detail="Invalid tenant_id format")
+
+    poi_repo = POIContextRepository(db)
+    deleted = await poi_repo.delete_by_tenant_id(tenant_uuid)
+
+    if not deleted:
+        raise HTTPException(
+            status_code=404,
+            detail=f"POI context not found for tenant {tenant_id}"
+        )
+
+    return {
+        "status": "success",
+        "message": f"POI context deleted for tenant {tenant_id}"
+    }
+
+
+@router.get("/{tenant_id}/feature-importance")
+async def get_feature_importance(
+    tenant_id: str,
+    db: AsyncSession = Depends(get_db)
+):
+    """
+    Get feature importance summary for tenant's POI context.
+
+    Shows which POI categories are relevant and their impact scores.
+    """
+    try:
+        tenant_uuid = uuid.UUID(tenant_id)
+    except ValueError:
+        raise HTTPException(status_code=400, detail="Invalid tenant_id format")
+
+    poi_repo = POIContextRepository(db)
+    poi_context = await poi_repo.get_by_tenant_id(tenant_uuid)
+
+    if not poi_context:
+        raise HTTPException(
+            status_code=404,
+            detail=f"POI context not found for tenant {tenant_id}"
+        )
+
+    feature_selector = POIFeatureSelector()
+    importance_summary = feature_selector.get_feature_importance_summary(
+        poi_context.poi_detection_results
+    )
+
+    return {
+        "tenant_id": tenant_id,
+        "feature_importance": importance_summary,
+        "total_categories": len(importance_summary),
+        "relevant_categories": sum(1 for cat in importance_summary if cat["is_relevant"])
+    }
+
+
+@router.get("/{tenant_id}/competitor-analysis")
+async def get_competitor_analysis(
+    tenant_id: str,
+    db: AsyncSession = Depends(get_db)
+):
+    """
+    Get detailed competitor analysis for tenant location.
+    """
+    try:
+        tenant_uuid = uuid.UUID(tenant_id)
+    except ValueError:
+        raise HTTPException(status_code=400, detail="Invalid tenant_id format")
+
+    poi_repo = POIContextRepository(db)
+    poi_context = await poi_repo.get_by_tenant_id(tenant_uuid)
+
+    if not poi_context:
+        raise HTTPException(
+            status_code=404,
+            detail=f"POI context not found for tenant {tenant_id}"
+        )
+
+    competitor_analyzer = CompetitorAnalyzer()
+    competitors = poi_context.poi_detection_results.get("competitors", {}).get("pois", [])
+
+    analysis = competitor_analyzer.analyze_competitive_landscape(
+        competitors,
+        (poi_context.latitude, poi_context.longitude),
+        tenant_id
+    )
+
+    insights = competitor_analyzer.get_competitive_insights(analysis)
+
+    return {
+        "tenant_id": tenant_id,
+        "location": {
+            "latitude": poi_context.latitude,
+            "longitude": poi_context.longitude
+        },
+        "competitor_analysis": analysis,
+        "insights": insights
+    }
+
+
+@router.get("/health")
+async def poi_health_check():
+    """
+    Check POI detection service health.
+
+    Verifies Overpass API accessibility.
+    """
+    poi_service = POIDetectionService()
+    health = await poi_service.health_check()
+
+    if not health["healthy"]:
+        raise HTTPException(
+            status_code=503,
+            detail=f"POI detection service unhealthy: {health.get('error', 'Unknown error')}"
+        )
+
+    return {
+        "status": "healthy",
+        "overpass_api": health
+    }
+
+
+@router.get("/cache/stats")
+async def get_cache_stats():
+    """
+    Get POI cache statistics.
+    """
+    try:
+        redis_client = await get_redis_client()
+        cache_service = POICacheService(redis_client)
+        stats = await cache_service.get_cache_stats()
+
+        return {
+            "status": "success",
+            "cache_stats": stats
+        }
+    except Exception as e:
+        logger.error("Failed to get cache stats", error=str(e))
+        raise HTTPException(
+            status_code=500,
+            detail=f"Failed to get cache stats: {str(e)}"
+        )

services/external/app/api/poi_refresh_jobs.py

@@ -0,0 +1,441 @@
+"""
+POI Refresh Jobs API Endpoints
+
+REST API for managing POI refresh background jobs.
+"""
+
+from fastapi import APIRouter, Depends, HTTPException, Query
+from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy import select, desc
+from typing import List, Optional
+from datetime import datetime, timezone
+from pydantic import BaseModel, Field
+import structlog
+import uuid
+
+from app.core.database import get_db
+from app.services.poi_refresh_service import POIRefreshService
+from app.services.poi_scheduler import get_scheduler
+from app.models.poi_refresh_job import POIRefreshJob
+
+logger = structlog.get_logger()
+
+router = APIRouter(prefix="/poi-refresh-jobs", tags=["POI Refresh Jobs"])
+
+
+# Response Models
+class POIRefreshJobResponse(BaseModel):
+    """POI refresh job response"""
+    id: str
+    tenant_id: str
+    status: str
+    scheduled_at: datetime
+    started_at: Optional[datetime] = None
+    completed_at: Optional[datetime] = None
+    attempt_count: int
+    max_attempts: int
+    pois_detected: Optional[int] = None
+    changes_detected: bool = False
+    change_summary: Optional[dict] = None
+    error_message: Optional[str] = None
+    next_scheduled_at: Optional[datetime] = None
+    duration_seconds: Optional[float] = None
+    is_overdue: bool
+    can_retry: bool
+
+    class Config:
+        from_attributes = True
+
+
+class ScheduleJobRequest(BaseModel):
+    """Schedule POI refresh job request"""
+    tenant_id: str = Field(..., description="Tenant UUID")
+    latitude: float = Field(..., ge=-90, le=90, description="Bakery latitude")
+    longitude: float = Field(..., ge=-180, le=180, description="Bakery longitude")
+    scheduled_at: Optional[datetime] = Field(None, description="When to run (default: 180 days from now)")
+
+
+class JobExecutionResult(BaseModel):
+    """Job execution result"""
+    status: str
+    job_id: str
+    message: Optional[str] = None
+    pois_detected: Optional[int] = None
+    changes_detected: Optional[bool] = None
+    change_summary: Optional[dict] = None
+    duration_seconds: Optional[float] = None
+    next_scheduled_at: Optional[str] = None
+    error: Optional[str] = None
+    attempt: Optional[int] = None
+    can_retry: Optional[bool] = None
+
+
+# Endpoints
+@router.post(
+    "/schedule",
+    response_model=POIRefreshJobResponse,
+    summary="Schedule POI refresh job",
+    description="Schedule a background job to refresh POI context for a tenant"
+)
+async def schedule_refresh_job(
+    request: ScheduleJobRequest,
+    db: AsyncSession = Depends(get_db)
+):
+    """
+    Schedule a POI refresh job for a tenant.
+
+    Creates a background job that will detect POIs for the tenant's location
+    at the scheduled time. Default schedule is 180 days from now.
+    """
+    try:
+        tenant_uuid = uuid.UUID(request.tenant_id)
+    except ValueError:
+        raise HTTPException(status_code=400, detail="Invalid tenant_id format")
+
+    try:
+        poi_refresh_service = POIRefreshService()
+        job = await poi_refresh_service.schedule_refresh_job(
+            tenant_id=request.tenant_id,
+            latitude=request.latitude,
+            longitude=request.longitude,
+            scheduled_at=request.scheduled_at,
+            session=db
+        )
+
+        logger.info(
+            "POI refresh job scheduled via API",
+            tenant_id=request.tenant_id,
+            job_id=str(job.id),
+            scheduled_at=job.scheduled_at
+        )
+
+        return POIRefreshJobResponse(
+            id=str(job.id),
+            tenant_id=str(job.tenant_id),
+            status=job.status,
+            scheduled_at=job.scheduled_at,
+            started_at=job.started_at,
+            completed_at=job.completed_at,
+            attempt_count=job.attempt_count,
+            max_attempts=job.max_attempts,
+            pois_detected=job.pois_detected,
+            changes_detected=job.changes_detected,
+            change_summary=job.change_summary,
+            error_message=job.error_message,
+            next_scheduled_at=job.next_scheduled_at,
+            duration_seconds=job.duration_seconds,
+            is_overdue=job.is_overdue,
+            can_retry=job.can_retry
+        )
+
+    except Exception as e:
+        logger.error(
+            "Failed to schedule POI refresh job",
+            tenant_id=request.tenant_id,
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"Failed to schedule refresh job: {str(e)}"
+        )
+
+
+@router.get(
+    "/{job_id}",
+    response_model=POIRefreshJobResponse,
+    summary="Get refresh job by ID",
+    description="Retrieve details of a specific POI refresh job"
+)
+async def get_refresh_job(
+    job_id: str,
+    db: AsyncSession = Depends(get_db)
+):
+    """Get POI refresh job by ID"""
+    try:
+        job_uuid = uuid.UUID(job_id)
+    except ValueError:
+        raise HTTPException(status_code=400, detail="Invalid job_id format")
+
+    result = await db.execute(
+        select(POIRefreshJob).where(POIRefreshJob.id == job_uuid)
+    )
+    job = result.scalar_one_or_none()
+
+    if not job:
+        raise HTTPException(status_code=404, detail=f"Job not found: {job_id}")
+
+    return POIRefreshJobResponse(
+        id=str(job.id),
+        tenant_id=str(job.tenant_id),
+        status=job.status,
+        scheduled_at=job.scheduled_at,
+        started_at=job.started_at,
+        completed_at=job.completed_at,
+        attempt_count=job.attempt_count,
+        max_attempts=job.max_attempts,
+        pois_detected=job.pois_detected,
+        changes_detected=job.changes_detected,
+        change_summary=job.change_summary,
+        error_message=job.error_message,
+        next_scheduled_at=job.next_scheduled_at,
+        duration_seconds=job.duration_seconds,
+        is_overdue=job.is_overdue,
+        can_retry=job.can_retry
+    )
+
+
+@router.get(
+    "/tenant/{tenant_id}",
+    response_model=List[POIRefreshJobResponse],
+    summary="Get refresh jobs for tenant",
+    description="Retrieve all POI refresh jobs for a specific tenant"
+)
+async def get_tenant_refresh_jobs(
+    tenant_id: str,
+    status: Optional[str] = Query(None, description="Filter by status"),
+    limit: int = Query(50, ge=1, le=200, description="Maximum number of results"),
+    db: AsyncSession = Depends(get_db)
+):
+    """Get all POI refresh jobs for a tenant"""
+    try:
+        tenant_uuid = uuid.UUID(tenant_id)
+    except ValueError:
+        raise HTTPException(status_code=400, detail="Invalid tenant_id format")
+
+    query = select(POIRefreshJob).where(POIRefreshJob.tenant_id == tenant_uuid)
+
+    if status:
+        query = query.where(POIRefreshJob.status == status)
+
+    query = query.order_by(desc(POIRefreshJob.scheduled_at)).limit(limit)
+
+    result = await db.execute(query)
+    jobs = result.scalars().all()
+
+    return [
+        POIRefreshJobResponse(
+            id=str(job.id),
+            tenant_id=str(job.tenant_id),
+            status=job.status,
+            scheduled_at=job.scheduled_at,
+            started_at=job.started_at,
+            completed_at=job.completed_at,
+            attempt_count=job.attempt_count,
+            max_attempts=job.max_attempts,
+            pois_detected=job.pois_detected,
+            changes_detected=job.changes_detected,
+            change_summary=job.change_summary,
+            error_message=job.error_message,
+            next_scheduled_at=job.next_scheduled_at,
+            duration_seconds=job.duration_seconds,
+            is_overdue=job.is_overdue,
+            can_retry=job.can_retry
+        )
+        for job in jobs
+    ]
+
+
+@router.post(
+    "/{job_id}/execute",
+    response_model=JobExecutionResult,
+    summary="Execute refresh job",
+    description="Manually trigger execution of a pending POI refresh job"
+)
+async def execute_refresh_job(
+    job_id: str,
+    db: AsyncSession = Depends(get_db)
+):
+    """Manually execute a POI refresh job"""
+    try:
+        job_uuid = uuid.UUID(job_id)
+    except ValueError:
+        raise HTTPException(status_code=400, detail="Invalid job_id format")
+
+    try:
+        poi_refresh_service = POIRefreshService()
+        result = await poi_refresh_service.execute_refresh_job(
+            job_id=job_id,
+            session=db
+        )
+
+        logger.info(
+            "POI refresh job executed via API",
+            job_id=job_id,
+            status=result["status"]
+        )
+
+        return JobExecutionResult(**result)
+
+    except ValueError as e:
+        raise HTTPException(status_code=404, detail=str(e))
+    except Exception as e:
+        logger.error(
+            "Failed to execute POI refresh job",
+            job_id=job_id,
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"Failed to execute refresh job: {str(e)}"
+        )
+
+
+@router.post(
+    "/process-pending",
+    summary="Process all pending jobs",
+    description="Manually trigger processing of all pending POI refresh jobs"
+)
+async def process_pending_jobs(
+    max_concurrent: int = Query(5, ge=1, le=20, description="Max concurrent executions"),
+    db: AsyncSession = Depends(get_db)
+):
+    """Process all pending POI refresh jobs"""
+    try:
+        poi_refresh_service = POIRefreshService()
+        result = await poi_refresh_service.process_pending_jobs(
+            max_concurrent=max_concurrent,
+            session=db
+        )
+
+        logger.info(
+            "Pending POI refresh jobs processed via API",
+            total_jobs=result["total_jobs"],
+            successful=result["successful"],
+            failed=result["failed"]
+        )
+
+        return result
+
+    except Exception as e:
+        logger.error(
+            "Failed to process pending POI refresh jobs",
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"Failed to process pending jobs: {str(e)}"
+        )
+
+
+@router.get(
+    "/pending",
+    response_model=List[POIRefreshJobResponse],
+    summary="Get pending jobs",
+    description="Retrieve all pending POI refresh jobs that are due for execution"
+)
+async def get_pending_jobs(
+    limit: int = Query(100, ge=1, le=500, description="Maximum number of results"),
+    db: AsyncSession = Depends(get_db)
+):
+    """Get all pending POI refresh jobs"""
+    try:
+        poi_refresh_service = POIRefreshService()
+        jobs = await poi_refresh_service.get_pending_jobs(
+            limit=limit,
+            session=db
+        )
+
+        return [
+            POIRefreshJobResponse(
+                id=str(job.id),
+                tenant_id=str(job.tenant_id),
+                status=job.status,
+                scheduled_at=job.scheduled_at,
+                started_at=job.started_at,
+                completed_at=job.completed_at,
+                attempt_count=job.attempt_count,
+                max_attempts=job.max_attempts,
+                pois_detected=job.pois_detected,
+                changes_detected=job.changes_detected,
+                change_summary=job.change_summary,
+                error_message=job.error_message,
+                next_scheduled_at=job.next_scheduled_at,
+                duration_seconds=job.duration_seconds,
+                is_overdue=job.is_overdue,
+                can_retry=job.can_retry
+            )
+            for job in jobs
+        ]
+
+    except Exception as e:
+        logger.error(
+            "Failed to get pending POI refresh jobs",
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"Failed to get pending jobs: {str(e)}"
+        )
+
+
+@router.post(
+    "/trigger-scheduler",
+    summary="Trigger scheduler immediately",
+    description="Trigger an immediate check for pending jobs (bypasses schedule)"
+)
+async def trigger_scheduler():
+    """Trigger POI refresh scheduler immediately"""
+    try:
+        scheduler = get_scheduler()
+
+        if not scheduler.is_running:
+            raise HTTPException(
+                status_code=503,
+                detail="POI refresh scheduler is not running"
+            )
+
+        result = await scheduler.trigger_immediate_check()
+
+        logger.info(
+            "POI refresh scheduler triggered via API",
+            total_jobs=result["total_jobs"],
+            successful=result["successful"],
+            failed=result["failed"]
+        )
+
+        return result
+
+    except HTTPException:
+        raise
+    except Exception as e:
+        logger.error(
+            "Failed to trigger POI refresh scheduler",
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"Failed to trigger scheduler: {str(e)}"
+        )
+
+
+@router.get(
+    "/scheduler/status",
+    summary="Get scheduler status",
+    description="Check if POI refresh scheduler is running"
+)
+async def get_scheduler_status():
+    """Get POI refresh scheduler status"""
+    try:
+        scheduler = get_scheduler()
+
+        return {
+            "is_running": scheduler.is_running,
+            "check_interval_seconds": scheduler.check_interval_seconds,
+            "max_concurrent_jobs": scheduler.max_concurrent_jobs
+        }
+
+    except Exception as e:
+        logger.error(
+            "Failed to get scheduler status",
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"Failed to get scheduler status: {str(e)}"
+        )

services/external/app/cache/poi_cache_service.py

@@ -0,0 +1,208 @@
+"""
+POI Cache Service
+
+Caches POI detection results to avoid hammering Overpass API.
+POI data doesn't change frequently, so aggressive caching is appropriate.
+"""
+
+from typing import Optional, Dict, Any
+import json
+import structlog
+from datetime import timedelta
+
+from app.core.poi_config import (
+    POI_CACHE_TTL_DAYS,
+    POI_COORDINATE_PRECISION
+)
+
+logger = structlog.get_logger()
+
+
+class POICacheService:
+    """
+    Redis-based cache for POI detection results.
+
+    Caches results by rounded coordinates to allow reuse for nearby locations.
+    Reduces load on Overpass API and improves onboarding performance.
+    """
+
+    def __init__(self, redis_client):
+        """
+        Initialize cache service.
+
+        Args:
+            redis_client: Redis client instance
+        """
+        self.redis = redis_client
+        self.cache_ttl_days = POI_CACHE_TTL_DAYS
+        self.coordinate_precision = POI_COORDINATE_PRECISION
+
+    def _generate_cache_key(self, latitude: float, longitude: float) -> str:
+        """
+        Generate cache key from coordinates.
+
+        Rounds coordinates to specified precision (default 4 decimals ≈ 10m).
+        This allows cache reuse for bakeries in very close proximity.
+
+        Args:
+            latitude: Bakery latitude
+            longitude: Bakery longitude
+
+        Returns:
+            Redis cache key
+        """
+        lat_rounded = round(latitude, self.coordinate_precision)
+        lon_rounded = round(longitude, self.coordinate_precision)
+        return f"poi_cache:{lat_rounded}:{lon_rounded}"
+
+    async def get_cached_pois(
+        self,
+        latitude: float,
+        longitude: float
+    ) -> Optional[Dict[str, Any]]:
+        """
+        Get cached POI results for location.
+
+        Args:
+            latitude: Bakery latitude
+            longitude: Bakery longitude
+
+        Returns:
+            Cached POI detection results or None if not cached
+        """
+        cache_key = self._generate_cache_key(latitude, longitude)
+
+        try:
+            cached_data = await self.redis.get(cache_key)
+            if cached_data:
+                logger.info(
+                    "POI cache hit",
+                    cache_key=cache_key,
+                    location=(latitude, longitude)
+                )
+                return json.loads(cached_data)
+            else:
+                logger.debug(
+                    "POI cache miss",
+                    cache_key=cache_key,
+                    location=(latitude, longitude)
+                )
+                return None
+        except Exception as e:
+            logger.warning(
+                "Failed to retrieve POI cache",
+                error=str(e),
+                cache_key=cache_key
+            )
+            return None
+
+    async def cache_poi_results(
+        self,
+        latitude: float,
+        longitude: float,
+        poi_data: Dict[str, Any]
+    ) -> bool:
+        """
+        Cache POI detection results.
+
+        Args:
+            latitude: Bakery latitude
+            longitude: Bakery longitude
+            poi_data: Complete POI detection results
+
+        Returns:
+            True if cached successfully, False otherwise
+        """
+        cache_key = self._generate_cache_key(latitude, longitude)
+        ttl_seconds = self.cache_ttl_days * 24 * 60 * 60
+
+        try:
+            await self.redis.setex(
+                cache_key,
+                ttl_seconds,
+                json.dumps(poi_data)
+            )
+            logger.info(
+                "POI results cached",
+                cache_key=cache_key,
+                ttl_days=self.cache_ttl_days,
+                location=(latitude, longitude)
+            )
+            return True
+        except Exception as e:
+            logger.error(
+                "Failed to cache POI results",
+                error=str(e),
+                cache_key=cache_key
+            )
+            return False
+
+    async def invalidate_cache(
+        self,
+        latitude: float,
+        longitude: float
+    ) -> bool:
+        """
+        Invalidate cached POI results for location.
+
+        Useful for manual refresh or data corrections.
+
+        Args:
+            latitude: Bakery latitude
+            longitude: Bakery longitude
+
+        Returns:
+            True if invalidated successfully
+        """
+        cache_key = self._generate_cache_key(latitude, longitude)
+
+        try:
+            deleted = await self.redis.delete(cache_key)
+            if deleted:
+                logger.info(
+                    "POI cache invalidated",
+                    cache_key=cache_key,
+                    location=(latitude, longitude)
+                )
+            return bool(deleted)
+        except Exception as e:
+            logger.error(
+                "Failed to invalidate POI cache",
+                error=str(e),
+                cache_key=cache_key
+            )
+            return False
+
+    async def get_cache_stats(self) -> Dict[str, Any]:
+        """
+        Get cache statistics.
+
+        Returns:
+            Dictionary with cache stats (key count, memory usage, etc.)
+        """
+        try:
+            # Count POI cache keys
+            pattern = "poi_cache:*"
+            cursor = 0
+            key_count = 0
+
+            while True:
+                cursor, keys = await self.redis.scan(
+                    cursor=cursor,
+                    match=pattern,
+                    count=100
+                )
+                key_count += len(keys)
+                if cursor == 0:
+                    break
+
+            return {
+                "total_cached_locations": key_count,
+                "cache_ttl_days": self.cache_ttl_days,
+                "coordinate_precision": self.coordinate_precision
+            }
+        except Exception as e:
+            logger.error("Failed to get cache stats", error=str(e))
+            return {
+                "error": str(e)
+            }

services/external/app/core/poi_config.py

@@ -0,0 +1,181 @@
+"""
+POI Detection Configuration
+
+Defines POI categories, search parameters, and relevance thresholds
+for automated Point of Interest detection and feature engineering.
+"""
+
+from dataclasses import dataclass
+from typing import Dict
+
+
+@dataclass
+class POICategory:
+    """POI category definition with OSM query and ML parameters"""
+    name: str
+    osm_query: str
+    search_radius_m: int
+    weight: float  # Importance weight for ML model (positive or negative)
+    description: str
+
+
+# POI Category Definitions based on OpenStreetMap tags
+# Research-based search radii and weights for bakery demand forecasting
+POI_CATEGORIES: Dict[str, POICategory] = {
+    "schools": POICategory(
+        name="schools",
+        osm_query='["amenity"~"school|kindergarten|university|college"]',
+        search_radius_m=500,
+        weight=1.5,  # High positive impact - morning drop-off rush
+        description="Educational institutions causing morning/afternoon rush patterns"
+    ),
+    "offices": POICategory(
+        name="offices",
+        osm_query='["office"]',
+        search_radius_m=800,
+        weight=1.3,  # Positive impact - weekday lunch/breakfast demand
+        description="Office buildings and business centers"
+    ),
+    "gyms_sports": POICategory(
+        name="gyms_sports",
+        osm_query='["leisure"~"fitness_centre|sports_centre|stadium"]',
+        search_radius_m=600,
+        weight=0.8,  # Moderate impact - morning/evening activity
+        description="Fitness centers and sports facilities"
+    ),
+    "residential": POICategory(
+        name="residential",
+        osm_query='["building"~"residential|apartments|house"]',
+        search_radius_m=400,
+        weight=1.0,  # Base demand from residents
+        description="Residential buildings and housing"
+    ),
+    "tourism": POICategory(
+        name="tourism",
+        osm_query='["tourism"~"attraction|museum|hotel|hostel|guest_house"]',
+        search_radius_m=1000,
+        weight=1.2,  # Positive impact - tourist foot traffic
+        description="Tourist attractions, hotels, and points of interest"
+    ),
+    "competitors": POICategory(
+        name="competitors",
+        osm_query='["shop"~"bakery|pastry|confectionery"]',
+        search_radius_m=1000,
+        weight=-0.5,  # Negative impact - competition pressure
+        description="Competing bakeries and pastry shops"
+    ),
+    "transport_hubs": POICategory(
+        name="transport_hubs",
+        osm_query='["public_transport"~"station|stop"]["railway"~"station|subway_entrance|tram_stop"]',
+        search_radius_m=800,
+        weight=1.4,  # High impact - commuter foot traffic
+        description="Public transport stations and hubs"
+    ),
+    "coworking": POICategory(
+        name="coworking",
+        osm_query='["amenity"="coworking_space"]',
+        search_radius_m=600,
+        weight=1.1,  # Moderate-high impact - flexible workers
+        description="Coworking spaces and shared offices"
+    ),
+    "retail": POICategory(
+        name="retail",
+        osm_query='["shop"]',
+        search_radius_m=500,
+        weight=0.9,  # Moderate impact - general foot traffic
+        description="Retail shops and commercial areas"
+    )
+}
+
+
+# Feature Relevance Thresholds
+# Determines which POI features are significant enough to include in ML models
+# Based on retail gravity model research and distance decay patterns
+RELEVANCE_THRESHOLDS: Dict[str, Dict[str, float]] = {
+    "schools": {
+        "min_proximity_score": 0.5,  # At least moderate proximity required
+        "max_distance_to_nearest_m": 500,  # Must be within 500m
+        "min_count": 1  # At least 1 school
+    },
+    "offices": {
+        "min_proximity_score": 0.3,
+        "max_distance_to_nearest_m": 800,
+        "min_count": 2  # Offices are common; need multiple for impact
+    },
+    "gyms_sports": {
+        "min_proximity_score": 0.4,
+        "max_distance_to_nearest_m": 600,
+        "min_count": 1
+    },
+    "residential": {
+        "min_proximity_score": 1.0,  # High threshold; residential is everywhere in cities
+        "max_distance_to_nearest_m": 400,
+        "min_count": 5  # Need significant residential density
+    },
+    "tourism": {
+        "min_proximity_score": 0.2,  # Lower threshold; tourism is high-impact even at distance
+        "max_distance_to_nearest_m": 1000,
+        "min_count": 1
+    },
+    "competitors": {
+        "min_proximity_score": 0.1,  # Any competition is relevant (even distant)
+        "max_distance_to_nearest_m": 1000,
+        "min_count": 1
+    },
+    "transport_hubs": {
+        "min_proximity_score": 0.4,
+        "max_distance_to_nearest_m": 800,
+        "min_count": 1
+    },
+    "coworking": {
+        "min_proximity_score": 0.3,
+        "max_distance_to_nearest_m": 600,
+        "min_count": 1
+    },
+    "retail": {
+        "min_proximity_score": 0.8,  # Retail is common; higher bar for relevance
+        "max_distance_to_nearest_m": 500,
+        "min_count": 3
+    }
+}
+
+
+# Overpass API Configuration
+OVERPASS_API_URL = "https://overpass-api.de/api/interpreter"
+OVERPASS_TIMEOUT_SECONDS = 30
+OVERPASS_MAX_RETRIES = 4  # Increased from 3 to 4 for better resilience
+OVERPASS_RETRY_DELAY_SECONDS = 2  # Base delay (will use exponential backoff)
+
+
+# POI Cache Configuration
+POI_CACHE_TTL_DAYS = 90  # Cache POI results for 90 days
+POI_REFRESH_INTERVAL_DAYS = 180  # Refresh every 6 months
+POI_COORDINATE_PRECISION = 4  # Decimal places for cache key (≈10m precision)
+
+
+# Distance Bands for Feature Engineering (meters)
+DISTANCE_BANDS = [
+    (0, 100),      # Immediate proximity
+    (100, 300),    # Primary catchment (walking distance)
+    (300, 500),    # Secondary catchment
+    (500, 1000)    # Tertiary catchment
+]
+
+
+# Competitive Pressure Zones
+COMPETITOR_ZONES = {
+    "direct": {
+        "max_distance_m": 100,
+        "pressure_multiplier": -1.0  # Strong negative impact
+    },
+    "nearby": {
+        "max_distance_m": 500,
+        "pressure_multiplier": -0.5  # Moderate negative impact
+    },
+    "market": {
+        "max_distance_m": 1000,
+        "min_count_for_district": 5,  # If 5+ bakeries = bakery district
+        "district_multiplier": 0.3,  # Positive impact (destination area)
+        "normal_multiplier": -0.2  # Slight negative (competitive market)
+    }
+}

services/external/app/core/redis_client.py

@@ -0,0 +1,16 @@
+"""
+Redis Client for POI Service
+
+Provides access to shared Redis client for POI caching.
+"""
+
+from shared.redis_utils import get_redis_client as get_shared_redis_client
+
+
+async def get_redis_client():
+    """
+    Get Redis client for POI service.
+
+    Uses shared Redis infrastructure from shared utilities.
+    """
+    return await get_shared_redis_client()

services/external/app/external/apis/madrid_traffic_client.py

@@ -66,7 +66,7 @@ class MadridTrafficClient(BaseTrafficClient, BaseAPIClient):
             # Parse XML data
             traffic_points = self.processor.parse_traffic_xml(xml_content)
             if not traffic_points:
-                self.logger.warning("No traffic points found in XML")
+                self.logger.warning("No traffic points found in XML - API may be temporarily unavailable")
                 return None
             
             # Find nearest traffic point

services/external/app/ingestion/adapters/madrid_adapter.py

@@ -98,32 +98,53 @@ class MadridAdapter(CityDataAdapter):
     async def validate_connection(self) -> bool:
         """Validate connection to AEMET and Madrid OpenData
 
-        Note: Validation is lenient - passes if traffic API works.
+        Note: Validation is lenient - allows partial failures for temporary API issues.
         AEMET rate limits may cause weather validation to fail during initialization.
+        Madrid traffic API outages should not block validation entirely.
         """
         try:
-            test_traffic = await self.traffic_client.get_current_traffic(
-                self.madrid_lat,
-                self.madrid_lon
-            )
+            traffic_validation_passed = False
+            weather_validation_passed = False
 
-            # Traffic API must work (critical for operations)
-            if test_traffic is None:
-                logger.error("Traffic API validation failed - this is critical")
+            # Try traffic API first
+            try:
+                test_traffic = await self.traffic_client.get_current_traffic(
+                    self.madrid_lat,
+                    self.madrid_lon
+                )
+
+                if test_traffic is not None and len(test_traffic) > 0:
+                    traffic_validation_passed = True
+                    logger.info("Traffic API validation successful")
+                else:
+                    logger.warning("Traffic API validation failed - temporary unavailability (proceeding anyway)")
+            except Exception as traffic_error:
+                logger.warning("Traffic API validation error (temporary unavailability) - proceeding anyway", error=str(traffic_error))
+
+            # Try weather API
+            try:
+                test_weather = await self.aemet_client.get_current_weather(
+                    self.madrid_lat,
+                    self.madrid_lon
+                )
+
+                if test_weather is not None:
+                    weather_validation_passed = True
+                    logger.info("Weather API validation successful")
+                else:
+                    logger.warning("Weather API validation failed (likely rate limited) - proceeding anyway")
+            except Exception as weather_error:
+                logger.warning("Weather API validation error - proceeding anyway", error=str(weather_error))
+
+            # At least one validation should pass for basic connectivity
+            if not traffic_validation_passed and not weather_validation_passed:
+                logger.error("Both traffic and weather API validations failed - no connectivity")
                 return False
 
-            # Try weather API, but don't fail validation if rate limited
-            test_weather = await self.aemet_client.get_current_weather(
-                self.madrid_lat,
-                self.madrid_lon
-            )
-
-            if test_weather is None:
-                logger.warning("Weather API validation failed (likely rate limited) - proceeding anyway")
-            else:
-                logger.info("Weather API validation successful")
-
-            # Pass validation if traffic works (weather can be fetched later)
+            # Return success if at least one API is accessible
+            logger.info("Adapter connection validation passed", 
+                       traffic_valid=traffic_validation_passed,
+                       weather_valid=weather_validation_passed)
             return True
 
         except Exception as e:

services/external/app/ingestion/ingestion_manager.py

@@ -58,7 +58,17 @@ class DataIngestionManager:
             failures=failures
         )
 
-        return successes == len(results)
+        # Consider success if we have at least some cities initialized (majority success)
+        # This allows the system to continue even if some external APIs are temporarily unavailable
+        if successes > 0:
+            logger.info(
+                "Partial success achieved - continuing with available data",
+                success_ratio=f"{successes}/{len(results)}"
+            )
+            return True
+        else:
+            logger.error("All city initializations failed - system cannot proceed")
+            return False
 
     async def initialize_city(
         self,
@@ -123,23 +133,55 @@ class DataIngestionManager:
                 logger.error("Adapter validation failed", city=city.name)
                 return False
 
-            weather_data = await adapter.fetch_historical_weather(
-                start_date, end_date
-            )
+            # Fetch data with error handling to allow partial success
+            weather_data = []
+            traffic_data = []
+            
+            # Fetch weather data
+            try:
+                weather_data = await adapter.fetch_historical_weather(
+                    start_date, end_date
+                )
+                logger.info("Weather data fetched successfully", 
+                           records=len(weather_data), city=city.name)
+            except Exception as weather_error:
+                logger.error("Failed to fetch weather data", 
+                           city=city.name, error=str(weather_error))
+                # Don't return False here - continue with whatever data we can get
 
-            traffic_data = await adapter.fetch_historical_traffic(
-                start_date, end_date
-            )
+            # Fetch traffic data
+            try:
+                traffic_data = await adapter.fetch_historical_traffic(
+                    start_date, end_date
+                )
+                logger.info("Traffic data fetched successfully", 
+                           records=len(traffic_data), city=city.name)
+            except Exception as traffic_error:
+                logger.error("Failed to fetch traffic data", 
+                           city=city.name, error=str(traffic_error))
+                # Don't return False here - continue with weather data only if available
 
+            # Store available data (at least one type should be available for partial success)
             async with self.database_manager.get_session() as session:
                 repo = CityDataRepository(session)
 
-                weather_stored = await repo.bulk_store_weather(
-                    city_id, weather_data
-                )
-                traffic_stored = await repo.bulk_store_traffic(
-                    city_id, traffic_data
-                )
+                weather_stored = 0
+                traffic_stored = 0
+
+                if weather_data:
+                    weather_stored = await repo.bulk_store_weather(
+                        city_id, weather_data
+                    )
+                
+                if traffic_data:
+                    traffic_stored = await repo.bulk_store_traffic(
+                        city_id, traffic_data
+                    )
+
+            # Only fail if both data types failed to fetch
+            if not weather_data and not traffic_data:
+                logger.error("Both weather and traffic data fetch failed", city=city.name)
+                return False
 
             logger.info(
                 "City initialization complete",

services/external/app/jobs/initialize_data.py

@@ -29,18 +29,21 @@ async def main(months: int = 24):
         logger.info("Proceeding to seed school calendars...")
         calendar_success = await manager.seed_school_calendars()
 
-        # Both must succeed
-        overall_success = weather_traffic_success and calendar_success
+        # Calendar seeding is critical, but weather/traffic can have partial success
+        overall_success = calendar_success and weather_traffic_success
 
         if overall_success:
             logger.info("✅ Data initialization completed successfully (weather, traffic, calendars)")
             sys.exit(0)
         else:
-            if not weather_traffic_success:
-                logger.error("❌ Weather/traffic initialization failed")
             if not calendar_success:
-                logger.error("❌ Calendar seeding failed")
-            sys.exit(1)
+                logger.error("❌ Calendar seeding failed - this is critical")
+                sys.exit(1)
+            elif not weather_traffic_success:
+                # Log as warning instead of error if some data was retrieved
+                logger.warning("⚠️  Weather/traffic initialization had partial failures, but system can continue")
+                logger.info("✅ Calendar seeding completed - system can operate with available data")
+                sys.exit(0)  # Allow partial success for weather/traffic
 
     except Exception as e:
         logger.error("❌ Fatal error during initialization", error=str(e))

services/external/app/main.py

@@ -9,8 +9,10 @@ from app.core.config import settings
 from app.core.database import database_manager
 from app.services.messaging import setup_messaging, cleanup_messaging
 from shared.service_base import StandardFastAPIService
+from shared.redis_utils import initialize_redis, close_redis
 # Include routers
-from app.api import weather_data, traffic_data, city_operations, calendar_operations, audit
+from app.api import weather_data, traffic_data, city_operations, calendar_operations, audit, poi_context, geocoding, poi_refresh_jobs
+from app.services.poi_scheduler import start_scheduler, stop_scheduler
 
 
 class ExternalService(StandardFastAPIService):
@@ -41,7 +43,8 @@ class ExternalService(StandardFastAPIService):
         # Define expected database tables for health checks
         external_expected_tables = [
             'weather_data', 'weather_forecasts', 'traffic_data',
-            'traffic_measurement_points', 'traffic_background_jobs'
+            'traffic_measurement_points', 'traffic_background_jobs',
+            'tenant_poi_contexts', 'poi_refresh_jobs'
         ]
 
         # Define custom API checks
@@ -146,12 +149,25 @@ class ExternalService(StandardFastAPIService):
 
     async def on_startup(self, app: FastAPI):
         """Custom startup logic for external service"""
-        pass
+        # Initialize Redis connection
+        await initialize_redis(settings.REDIS_URL, db=0, max_connections=50)
+        self.logger.info("Redis initialized for external service")
+
+        # Start POI refresh scheduler
+        await start_scheduler()
+        self.logger.info("POI refresh scheduler started")
 
     async def on_shutdown(self, app: FastAPI):
         """Custom shutdown logic for external service"""
+        # Stop POI refresh scheduler
+        await stop_scheduler()
+        self.logger.info("POI refresh scheduler stopped")
+        
+        # Close Redis connection
+        await close_redis()
+        self.logger.info("Redis connection closed")
+        
         # Database cleanup is handled by the base class
-        pass
 
     def get_service_features(self):
         """Return external-specific features"""
@@ -182,4 +198,7 @@ service.add_router(audit.router)
 service.add_router(weather_data.router)
 service.add_router(traffic_data.router)
 service.add_router(city_operations.router)  # New v2.0 city-based optimized endpoints
-service.add_router(calendar_operations.router)  # School calendars and hyperlocal data
+service.add_router(calendar_operations.router)  # School calendars and hyperlocal data
+service.add_router(poi_context.router)  # POI detection and location-based features
+service.add_router(geocoding.router)  # Address search and geocoding
+service.add_router(poi_refresh_jobs.router)  # POI refresh background jobs

services/external/app/models/poi_context.py

@@ -0,0 +1,123 @@
+"""
+POI Context Model
+
+Stores Point of Interest detection results and ML features for bakery locations.
+Used for location-based demand forecasting with contextual features.
+"""
+
+from sqlalchemy import Column, String, DateTime, Float, Index, Integer
+from sqlalchemy.dialects.postgresql import UUID, JSONB
+from datetime import datetime, timezone, timedelta
+import uuid
+
+from app.core.database import Base
+
+
+class TenantPOIContext(Base):
+    """
+    POI (Point of Interest) context for bakery location.
+
+    Stores detected POIs around bakery and calculated ML features
+    for demand forecasting with location-specific context.
+    """
+
+    __tablename__ = "tenant_poi_contexts"
+
+    id = Column(UUID(as_uuid=True), primary_key=True, default=uuid.uuid4)
+    tenant_id = Column(UUID(as_uuid=True), nullable=False, unique=True, index=True)
+
+    # Location (denormalized for quick reference and spatial queries)
+    latitude = Column(Float, nullable=False)
+    longitude = Column(Float, nullable=False)
+
+    # POI Detection Results (full raw data)
+    # Structure: {
+    #   "schools": {
+    #     "pois": [{"osm_id": "...", "name": "...", "lat": ..., "lon": ...}],
+    #     "features": {"proximity_score": 3.45, "count_0_100m": 2, ...},
+    #     "count": 5
+    #   },
+    #   "offices": {...},
+    #   ...
+    # }
+    poi_detection_results = Column(JSONB, nullable=False, default=dict)
+
+    # ML Features (flat structure for easy model ingestion)
+    # Structure: {
+    #   "poi_schools_proximity_score": 3.45,
+    #   "poi_schools_weighted_proximity_score": 5.18,
+    #   "poi_schools_count_0_100m": 2,
+    #   "poi_offices_proximity_score": 1.23,
+    #   ...
+    # }
+    ml_features = Column(JSONB, nullable=False, default=dict)
+
+    # Summary Statistics
+    total_pois_detected = Column(Integer, default=0)
+    high_impact_categories = Column(JSONB, default=list)  # Categories with significant POI presence
+    relevant_categories = Column(JSONB, default=list)  # Categories that passed relevance thresholds
+
+    # Detection Metadata
+    detection_timestamp = Column(DateTime(timezone=True), nullable=False)
+    detection_source = Column(String(50), default="overpass_api")
+    detection_status = Column(String(20), default="completed")  # completed, failed, partial
+    detection_error = Column(String(500), nullable=True)  # Error message if detection failed
+
+    # Data Freshness Strategy
+    # POIs don't change frequently, refresh every 6 months
+    next_refresh_date = Column(DateTime(timezone=True), nullable=True)
+    refresh_interval_days = Column(Integer, default=180)  # 6 months default
+    last_refreshed_at = Column(DateTime(timezone=True), nullable=True)
+
+    # Timestamps
+    created_at = Column(DateTime(timezone=True), default=lambda: datetime.now(timezone.utc))
+    updated_at = Column(
+        DateTime(timezone=True),
+        default=lambda: datetime.now(timezone.utc),
+        onupdate=lambda: datetime.now(timezone.utc)
+    )
+
+    __table_args__ = (
+        Index('idx_tenant_poi_location', 'latitude', 'longitude'),
+        Index('idx_tenant_poi_refresh', 'next_refresh_date'),
+        Index('idx_tenant_poi_status', 'detection_status'),
+    )
+
+    def to_dict(self):
+        """Convert to dictionary for API responses"""
+        return {
+            "id": str(self.id),
+            "tenant_id": str(self.tenant_id),
+            "location": {
+                "latitude": self.latitude,
+                "longitude": self.longitude
+            },
+            "poi_detection_results": self.poi_detection_results,
+            "ml_features": self.ml_features,
+            "total_pois_detected": self.total_pois_detected,
+            "high_impact_categories": self.high_impact_categories,
+            "relevant_categories": self.relevant_categories,
+            "detection_timestamp": self.detection_timestamp.isoformat() if self.detection_timestamp else None,
+            "detection_source": self.detection_source,
+            "detection_status": self.detection_status,
+            "detection_error": self.detection_error,
+            "next_refresh_date": self.next_refresh_date.isoformat() if self.next_refresh_date else None,
+            "last_refreshed_at": self.last_refreshed_at.isoformat() if self.last_refreshed_at else None,
+            "created_at": self.created_at.isoformat() if self.created_at else None,
+            "updated_at": self.updated_at.isoformat() if self.updated_at else None
+        }
+
+    def is_stale(self) -> bool:
+        """Check if POI data needs refresh"""
+        if not self.next_refresh_date:
+            return True
+        return datetime.now(timezone.utc) > self.next_refresh_date
+
+    def calculate_next_refresh(self) -> datetime:
+        """Calculate next refresh date based on interval"""
+        return datetime.now(timezone.utc) + timedelta(days=self.refresh_interval_days)
+
+    def mark_refreshed(self):
+        """Mark as refreshed and calculate next refresh date"""
+        self.last_refreshed_at = datetime.now(timezone.utc)
+        self.next_refresh_date = self.calculate_next_refresh()

services/external/app/models/poi_refresh_job.py

@@ -0,0 +1,154 @@
+"""
+POI Refresh Job Model
+
+Tracks background jobs for periodic POI context refresh.
+"""
+
+from sqlalchemy import Column, String, DateTime, Integer, Boolean, Text, Float
+from sqlalchemy.dialects.postgresql import UUID, JSONB
+from datetime import datetime, timezone
+import uuid
+
+from app.core.database import Base
+
+
+class POIRefreshJob(Base):
+    """
+    POI Refresh Background Job Model
+
+    Tracks periodic POI context refresh jobs for all tenants.
+    Jobs run on a configurable schedule (default: 180 days).
+    """
+
+    __tablename__ = "poi_refresh_jobs"
+
+    id = Column(UUID(as_uuid=True), primary_key=True, default=uuid.uuid4)
+    tenant_id = Column(UUID(as_uuid=True), nullable=False, index=True)
+
+    # Job scheduling
+    scheduled_at = Column(
+        DateTime(timezone=True),
+        nullable=False,
+        index=True,
+        comment="When this job was scheduled"
+    )
+    started_at = Column(
+        DateTime(timezone=True),
+        nullable=True,
+        comment="When job execution started"
+    )
+    completed_at = Column(
+        DateTime(timezone=True),
+        nullable=True,
+        comment="When job execution completed"
+    )
+
+    # Job status
+    status = Column(
+        String(50),
+        nullable=False,
+        default="pending",
+        index=True,
+        comment="Job status: pending, running, completed, failed"
+    )
+
+    # Job execution details
+    attempt_count = Column(
+        Integer,
+        nullable=False,
+        default=0,
+        comment="Number of execution attempts"
+    )
+    max_attempts = Column(
+        Integer,
+        nullable=False,
+        default=3,
+        comment="Maximum number of retry attempts"
+    )
+
+    # Location data (cached for job execution)
+    latitude = Column(
+        Float,
+        nullable=False,
+        comment="Bakery latitude for POI detection"
+    )
+    longitude = Column(
+        Float,
+        nullable=False,
+        comment="Bakery longitude for POI detection"
+    )
+
+    # Results
+    pois_detected = Column(
+        Integer,
+        nullable=True,
+        comment="Number of POIs detected in this refresh"
+    )
+    changes_detected = Column(
+        Boolean,
+        default=False,
+        comment="Whether significant changes were detected"
+    )
+    change_summary = Column(
+        JSONB,
+        nullable=True,
+        comment="Summary of changes detected"
+    )
+
+    # Error handling
+    error_message = Column(
+        Text,
+        nullable=True,
+        comment="Error message if job failed"
+    )
+    error_details = Column(
+        JSONB,
+        nullable=True,
+        comment="Detailed error information"
+    )
+
+    # Next execution
+    next_scheduled_at = Column(
+        DateTime(timezone=True),
+        nullable=True,
+        index=True,
+        comment="When next refresh should be scheduled"
+    )
+
+    # Metadata
+    created_at = Column(
+        DateTime(timezone=True),
+        nullable=False,
+        default=lambda: datetime.now(timezone.utc)
+    )
+    updated_at = Column(
+        DateTime(timezone=True),
+        nullable=False,
+        default=lambda: datetime.now(timezone.utc),
+        onupdate=lambda: datetime.now(timezone.utc)
+    )
+
+    def __repr__(self):
+        return (
+            f"<POIRefreshJob(id={self.id}, tenant_id={self.tenant_id}, "
+            f"status={self.status}, scheduled_at={self.scheduled_at})>"
+        )
+
+    @property
+    def is_overdue(self) -> bool:
+        """Check if job is overdue for execution"""
+        if self.status in ("completed", "running"):
+            return False
+        return datetime.now(timezone.utc) > self.scheduled_at
+
+    @property
+    def can_retry(self) -> bool:
+        """Check if job can be retried"""
+        return self.attempt_count < self.max_attempts
+
+    @property
+    def duration_seconds(self) -> float | None:
+        """Calculate job duration in seconds"""
+        if self.started_at and self.completed_at:
+            return (self.completed_at - self.started_at).total_seconds()
+        return None

services/external/app/repositories/poi_context_repository.py

@@ -0,0 +1,271 @@
+"""
+POI Context Repository
+
+Data access layer for TenantPOIContext model.
+Handles CRUD operations for POI detection results and ML features.
+"""
+
+from typing import Optional, List
+from datetime import datetime, timezone
+from sqlalchemy import select, update, delete
+from sqlalchemy.ext.asyncio import AsyncSession
+import structlog
+import uuid
+
+from app.models.poi_context import TenantPOIContext
+
+logger = structlog.get_logger()
+
+
+class POIContextRepository:
+    """
+    Repository for POI context data access.
+
+    Manages storage and retrieval of POI detection results
+    and ML features for tenant locations.
+    """
+
+    def __init__(self, session: AsyncSession):
+        """
+        Initialize repository.
+
+        Args:
+            session: SQLAlchemy async session
+        """
+        self.session = session
+
+    async def create(self, poi_context_data: dict) -> TenantPOIContext:
+        """
+        Create new POI context record.
+
+        Args:
+            poi_context_data: Dictionary with POI context data
+
+        Returns:
+            Created TenantPOIContext instance
+        """
+        poi_context = TenantPOIContext(
+            tenant_id=poi_context_data["tenant_id"],
+            latitude=poi_context_data["latitude"],
+            longitude=poi_context_data["longitude"],
+            poi_detection_results=poi_context_data.get("poi_detection_results", {}),
+            ml_features=poi_context_data.get("ml_features", {}),
+            total_pois_detected=poi_context_data.get("total_pois_detected", 0),
+            high_impact_categories=poi_context_data.get("high_impact_categories", []),
+            relevant_categories=poi_context_data.get("relevant_categories", []),
+            detection_timestamp=poi_context_data.get(
+                "detection_timestamp",
+                datetime.now(timezone.utc)
+            ),
+            detection_source=poi_context_data.get("detection_source", "overpass_api"),
+            detection_status=poi_context_data.get("detection_status", "completed"),
+            detection_error=poi_context_data.get("detection_error"),
+            refresh_interval_days=poi_context_data.get("refresh_interval_days", 180)
+        )
+
+        # Calculate next refresh date
+        poi_context.next_refresh_date = poi_context.calculate_next_refresh()
+
+        self.session.add(poi_context)
+        await self.session.commit()
+        await self.session.refresh(poi_context)
+
+        logger.info(
+            "POI context created",
+            tenant_id=str(poi_context.tenant_id),
+            total_pois=poi_context.total_pois_detected
+        )
+
+        return poi_context
+
+    async def get_by_tenant_id(self, tenant_id: str | uuid.UUID) -> Optional[TenantPOIContext]:
+        """
+        Get POI context by tenant ID.
+
+        Args:
+            tenant_id: Tenant UUID
+
+        Returns:
+            TenantPOIContext or None if not found
+        """
+        if isinstance(tenant_id, str):
+            tenant_id = uuid.UUID(tenant_id)
+
+        stmt = select(TenantPOIContext).where(
+            TenantPOIContext.tenant_id == tenant_id
+        )
+        result = await self.session.execute(stmt)
+        return result.scalar_one_or_none()
+
+    async def get_by_id(self, poi_context_id: str | uuid.UUID) -> Optional[TenantPOIContext]:
+        """
+        Get POI context by ID.
+
+        Args:
+            poi_context_id: POI context UUID
+
+        Returns:
+            TenantPOIContext or None if not found
+        """
+        if isinstance(poi_context_id, str):
+            poi_context_id = uuid.UUID(poi_context_id)
+
+        stmt = select(TenantPOIContext).where(
+            TenantPOIContext.id == poi_context_id
+        )
+        result = await self.session.execute(stmt)
+        return result.scalar_one_or_none()
+
+    async def update(
+        self,
+        tenant_id: str | uuid.UUID,
+        update_data: dict
+    ) -> Optional[TenantPOIContext]:
+        """
+        Update POI context for tenant.
+
+        Args:
+            tenant_id: Tenant UUID
+            update_data: Dictionary with fields to update
+
+        Returns:
+            Updated TenantPOIContext or None if not found
+        """
+        if isinstance(tenant_id, str):
+            tenant_id = uuid.UUID(tenant_id)
+
+        poi_context = await self.get_by_tenant_id(tenant_id)
+        if not poi_context:
+            return None
+
+        # Update fields
+        for key, value in update_data.items():
+            if hasattr(poi_context, key):
+                setattr(poi_context, key, value)
+
+        # Update timestamp
+        poi_context.updated_at = datetime.now(timezone.utc)
+
+        await self.session.commit()
+        await self.session.refresh(poi_context)
+
+        logger.info(
+            "POI context updated",
+            tenant_id=str(tenant_id),
+            updated_fields=list(update_data.keys())
+        )
+
+        return poi_context
+
+    async def create_or_update(
+        self,
+        tenant_id: str | uuid.UUID,
+        poi_detection_results: dict
+    ) -> TenantPOIContext:
+        """
+        Create new POI context or update existing one.
+
+        Args:
+            tenant_id: Tenant UUID
+            poi_detection_results: Full POI detection results
+
+        Returns:
+            Created or updated TenantPOIContext
+        """
+        if isinstance(tenant_id, str):
+            tenant_id = uuid.UUID(tenant_id)
+
+        existing = await self.get_by_tenant_id(tenant_id)
+
+        poi_context_data = {
+            "tenant_id": tenant_id,
+            "latitude": poi_detection_results["location"]["latitude"],
+            "longitude": poi_detection_results["location"]["longitude"],
+            "poi_detection_results": poi_detection_results.get("poi_categories", {}),
+            "ml_features": poi_detection_results.get("ml_features", {}),
+            "total_pois_detected": poi_detection_results.get("summary", {}).get("total_pois_detected", 0),
+            "high_impact_categories": poi_detection_results.get("summary", {}).get("high_impact_categories", []),
+            "relevant_categories": poi_detection_results.get("relevant_categories", []),
+            "detection_timestamp": datetime.fromisoformat(
+                poi_detection_results["detection_timestamp"].replace("Z", "+00:00")
+            ) if isinstance(poi_detection_results.get("detection_timestamp"), str)
+            else datetime.now(timezone.utc),
+            "detection_status": poi_detection_results.get("detection_status", "completed"),
+            "detection_error": None if poi_detection_results.get("detection_status") == "completed"
+            else str(poi_detection_results.get("detection_errors"))
+        }
+
+        if existing:
+            # Update existing
+            update_data = {
+                **poi_context_data,
+                "last_refreshed_at": datetime.now(timezone.utc)
+            }
+            existing.mark_refreshed()  # Update next_refresh_date
+            return await self.update(tenant_id, update_data)
+        else:
+            # Create new
+            return await self.create(poi_context_data)
+
+    async def delete_by_tenant_id(self, tenant_id: str | uuid.UUID) -> bool:
+        """
+        Delete POI context for tenant.
+
+        Args:
+            tenant_id: Tenant UUID
+
+        Returns:
+            True if deleted, False if not found
+        """
+        if isinstance(tenant_id, str):
+            tenant_id = uuid.UUID(tenant_id)
+
+        stmt = delete(TenantPOIContext).where(
+            TenantPOIContext.tenant_id == tenant_id
+        )
+        result = await self.session.execute(stmt)
+        await self.session.commit()
+
+        deleted = result.rowcount > 0
+        if deleted:
+            logger.info("POI context deleted", tenant_id=str(tenant_id))
+
+        return deleted
+
+    async def get_stale_contexts(self, limit: int = 100) -> List[TenantPOIContext]:
+        """
+        Get POI contexts that need refresh.
+
+        Args:
+            limit: Maximum number of contexts to return
+
+        Returns:
+            List of stale TenantPOIContext instances
+        """
+        now = datetime.now(timezone.utc)
+        stmt = (
+            select(TenantPOIContext)
+            .where(TenantPOIContext.next_refresh_date <= now)
+            .limit(limit)
+        )
+        result = await self.session.execute(stmt)
+        return list(result.scalars().all())
+
+    async def count_by_status(self) -> dict:
+        """
+        Count POI contexts by detection status.
+
+        Returns:
+            Dictionary with counts by status
+        """
+        from sqlalchemy import func
+
+        stmt = select(
+            TenantPOIContext.detection_status,
+            func.count(TenantPOIContext.id)
+        ).group_by(TenantPOIContext.detection_status)
+
+        result = await self.session.execute(stmt)
+        rows = result.all()
+
+        return {status: count for status, count in rows}

services/external/app/services/competitor_analyzer.py

@@ -0,0 +1,269 @@
+"""
+Competitor Analyzer
+
+Specialized analysis for competitor bakeries with competitive pressure modeling.
+Treats competitor proximity differently than other POIs, considering market dynamics.
+"""
+
+from typing import Dict, List, Any, Tuple
+import structlog
+from math import radians, sin, cos, sqrt, atan2
+
+from app.core.poi_config import COMPETITOR_ZONES
+
+logger = structlog.get_logger()
+
+
+class CompetitorAnalyzer:
+    """
+    Competitive landscape analyzer for bakery locations.
+
+    Models competitive pressure considering:
+    - Direct competition (<100m): Strong negative impact
+    - Nearby competition (100-500m): Moderate negative impact
+    - Market saturation (500-1000m): Can be positive (bakery district)
+      or negative (competitive market)
+    """
+
+    def analyze_competitive_landscape(
+        self,
+        competitor_pois: List[Dict[str, Any]],
+        bakery_location: Tuple[float, float],
+        tenant_id: str = None
+    ) -> Dict[str, Any]:
+        """
+        Analyze competitive pressure from nearby bakeries.
+
+        Args:
+            competitor_pois: List of detected competitor POIs
+            bakery_location: Tuple of (latitude, longitude)
+            tenant_id: Optional tenant ID for logging
+
+        Returns:
+            Competitive analysis with pressure scores and market classification
+        """
+        if not competitor_pois:
+            logger.info(
+                "No competitors detected - underserved market",
+                tenant_id=tenant_id
+            )
+            return {
+                "competitive_pressure_score": 0.0,
+                "direct_competitors_count": 0,
+                "nearby_competitors_count": 0,
+                "market_competitors_count": 0,
+                "competitive_zone": "low_competition",
+                "market_type": "underserved",
+                "competitive_advantage": "first_mover",
+                "ml_feature_competitive_pressure": 0.0,
+                "ml_feature_has_direct_competitor": 0,
+                "ml_feature_competitor_density_500m": 0,
+                "competitor_details": []
+            }
+
+        # Categorize competitors by distance
+        direct_competitors = []  # <100m
+        nearby_competitors = []  # 100-500m
+        market_competitors = []  # 500-1000m
+        competitor_details = []
+
+        for poi in competitor_pois:
+            distance_m = self._calculate_distance(
+                bakery_location, (poi["lat"], poi["lon"])
+            ) * 1000
+
+            competitor_info = {
+                "name": poi.get("name", "Unnamed"),
+                "osm_id": poi.get("osm_id"),
+                "distance_m": round(distance_m, 1),
+                "lat": poi["lat"],
+                "lon": poi["lon"]
+            }
+
+            if distance_m < COMPETITOR_ZONES["direct"]["max_distance_m"]:
+                direct_competitors.append(poi)
+                competitor_info["zone"] = "direct"
+            elif distance_m < COMPETITOR_ZONES["nearby"]["max_distance_m"]:
+                nearby_competitors.append(poi)
+                competitor_info["zone"] = "nearby"
+            elif distance_m < COMPETITOR_ZONES["market"]["max_distance_m"]:
+                market_competitors.append(poi)
+                competitor_info["zone"] = "market"
+
+            competitor_details.append(competitor_info)
+
+        # Calculate competitive pressure score
+        direct_pressure = (
+            len(direct_competitors) *
+            COMPETITOR_ZONES["direct"]["pressure_multiplier"]
+        )
+        nearby_pressure = (
+            len(nearby_competitors) *
+            COMPETITOR_ZONES["nearby"]["pressure_multiplier"]
+        )
+
+        # Market saturation analysis
+        min_for_district = COMPETITOR_ZONES["market"]["min_count_for_district"]
+        if len(market_competitors) >= min_for_district:
+            # Many bakeries = destination area (bakery district)
+            market_pressure = COMPETITOR_ZONES["market"]["district_multiplier"]
+            market_type = "bakery_district"
+        elif len(market_competitors) > 2:
+            market_pressure = COMPETITOR_ZONES["market"]["normal_multiplier"]
+            market_type = "competitive_market"
+        else:
+            market_pressure = 0.0
+            market_type = "normal_market"
+
+        competitive_pressure_score = (
+            direct_pressure + nearby_pressure + market_pressure
+        )
+
+        # Determine competitive zone classification
+        if len(direct_competitors) > 0:
+            competitive_zone = "high_competition"
+            competitive_advantage = "differentiation_required"
+        elif len(nearby_competitors) > 2:
+            competitive_zone = "moderate_competition"
+            competitive_advantage = "quality_focused"
+        else:
+            competitive_zone = "low_competition"
+            competitive_advantage = "local_leader"
+
+        # Sort competitors by distance
+        competitor_details.sort(key=lambda x: x["distance_m"])
+
+        logger.info(
+            "Competitive analysis complete",
+            tenant_id=tenant_id,
+            competitive_zone=competitive_zone,
+            market_type=market_type,
+            total_competitors=len(competitor_pois),
+            direct=len(direct_competitors),
+            nearby=len(nearby_competitors),
+            market=len(market_competitors),
+            pressure_score=competitive_pressure_score
+        )
+
+        return {
+            # Summary scores
+            "competitive_pressure_score": round(competitive_pressure_score, 2),
+
+            # Competitor counts by zone
+            "direct_competitors_count": len(direct_competitors),
+            "nearby_competitors_count": len(nearby_competitors),
+            "market_competitors_count": len(market_competitors),
+            "total_competitors_count": len(competitor_pois),
+
+            # Market classification
+            "competitive_zone": competitive_zone,
+            "market_type": market_type,
+            "competitive_advantage": competitive_advantage,
+
+            # ML features (for model integration)
+            "ml_feature_competitive_pressure": round(competitive_pressure_score, 2),
+            "ml_feature_has_direct_competitor": 1 if len(direct_competitors) > 0 else 0,
+            "ml_feature_competitor_density_500m": (
+                len(direct_competitors) + len(nearby_competitors)
+            ),
+
+            # Detailed competitor information
+            "competitor_details": competitor_details,
+
+            # Nearest competitor
+            "nearest_competitor": competitor_details[0] if competitor_details else None
+        }
+
+    def _calculate_distance(
+        self,
+        coord1: Tuple[float, float],
+        coord2: Tuple[float, float]
+    ) -> float:
+        """
+        Calculate Haversine distance in kilometers.
+
+        Args:
+            coord1: Tuple of (latitude, longitude)
+            coord2: Tuple of (latitude, longitude)
+
+        Returns:
+            Distance in kilometers
+        """
+        lat1, lon1 = coord1
+        lat2, lon2 = coord2
+
+        R = 6371  # Earth radius in km
+
+        dlat = radians(lat2 - lat1)
+        dlon = radians(lon2 - lon1)
+
+        a = (sin(dlat/2)**2 +
+             cos(radians(lat1)) * cos(radians(lat2)) * sin(dlon/2)**2)
+        c = 2 * atan2(sqrt(a), sqrt(1-a))
+
+        return R * c
+
+    def get_competitive_insights(
+        self,
+        analysis_result: Dict[str, Any]
+    ) -> List[str]:
+        """
+        Generate human-readable competitive insights.
+
+        Args:
+            analysis_result: Result from analyze_competitive_landscape
+
+        Returns:
+            List of insight strings for business intelligence
+        """
+        insights = []
+
+        zone = analysis_result["competitive_zone"]
+        market = analysis_result["market_type"]
+        pressure = analysis_result["competitive_pressure_score"]
+        direct = analysis_result["direct_competitors_count"]
+        nearby = analysis_result["nearby_competitors_count"]
+
+        # Zone-specific insights
+        if zone == "high_competition":
+            insights.append(
+                f"⚠️ High competition: {direct} direct competitor(s) within 100m. "
+                "Focus on differentiation and quality."
+            )
+        elif zone == "moderate_competition":
+            insights.append(
+                f"Moderate competition: {nearby} nearby competitor(s) within 500m. "
+                "Good opportunity for market share."
+            )
+        else:
+            insights.append(
+                "✅ Low competition: Local market leader opportunity."
+            )
+
+        # Market type insights
+        if market == "bakery_district":
+            insights.append(
+                "📍 Bakery district: High foot traffic area with multiple bakeries. "
+                "Customers actively seek bakery products here."
+            )
+        elif market == "competitive_market":
+            insights.append(
+                "Market has multiple bakeries. Quality and customer service critical."
+            )
+        elif market == "underserved":
+            insights.append(
+                "🎯 Underserved market: Potential for strong customer base growth."
+            )
+
+        # Pressure score insight
+        if pressure < -1.5:
+            insights.append(
+                "Strong competitive pressure expected to impact demand. "
+                "Marketing and differentiation essential."
+            )
+        elif pressure > 0:
+            insights.append(
+                "Positive market dynamics: Location benefits from bakery destination traffic."
+            )
+
+        return insights

services/external/app/services/nominatim_service.py

@@ -0,0 +1,282 @@
+"""
+Nominatim Geocoding Service
+
+Provides address search and geocoding using OpenStreetMap Nominatim API.
+For development: uses public API (rate-limited)
+For production: should point to self-hosted Nominatim instance
+"""
+
+import httpx
+from typing import List, Dict, Any, Optional
+import structlog
+from asyncio import sleep
+
+logger = structlog.get_logger()
+
+
+class NominatimService:
+    """
+    Nominatim geocoding and address search service.
+
+    Uses OpenStreetMap Nominatim API for address autocomplete and geocoding.
+    Respects rate limits and usage policy.
+    """
+
+    # For development: public API (rate-limited to 1 req/sec)
+    # For production: should be overridden with self-hosted instance
+    DEFAULT_BASE_URL = "https://nominatim.openstreetmap.org"
+
+    def __init__(self, base_url: Optional[str] = None, user_agent: str = "BakeryIA-Forecasting/1.0"):
+        """
+        Initialize Nominatim service.
+
+        Args:
+            base_url: Nominatim server URL (defaults to public API)
+            user_agent: User agent for API requests (required by Nominatim policy)
+        """
+        self.base_url = (base_url or self.DEFAULT_BASE_URL).rstrip("/")
+        self.user_agent = user_agent
+        self.headers = {
+            "User-Agent": self.user_agent
+        }
+
+        # Rate limiting for public API (1 request per second)
+        self.is_public_api = self.base_url == self.DEFAULT_BASE_URL
+        self.min_request_interval = 1.0 if self.is_public_api else 0.0
+
+        logger.info(
+            "Nominatim service initialized",
+            base_url=self.base_url,
+            is_public_api=self.is_public_api,
+            rate_limit=f"{self.min_request_interval}s" if self.is_public_api else "none"
+        )
+
+    async def search_address(
+        self,
+        query: str,
+        country_code: str = "es",
+        limit: int = 10
+    ) -> List[Dict[str, Any]]:
+        """
+        Search for addresses matching query (autocomplete).
+
+        Args:
+            query: Address search query
+            country_code: ISO country code to restrict search (default: Spain)
+            limit: Maximum number of results
+
+        Returns:
+            List of address suggestions with display_name, lat, lon, osm_id, etc.
+        """
+        if not query or len(query.strip()) < 3:
+            logger.warning("Search query too short", query=query)
+            return []
+
+        try:
+            # Rate limiting for public API
+            if self.is_public_api:
+                await sleep(self.min_request_interval)
+
+            async with httpx.AsyncClient(timeout=10.0) as client:
+                response = await client.get(
+                    f"{self.base_url}/search",
+                    params={
+                        "q": query,
+                        "format": "json",
+                        "addressdetails": 1,
+                        "countrycodes": country_code,
+                        "limit": limit,
+                        "accept-language": "es"
+                    },
+                    headers=self.headers
+                )
+                response.raise_for_status()
+                results = response.json()
+
+                # Parse and enrich results
+                addresses = []
+                for result in results:
+                    addresses.append({
+                        "display_name": result.get("display_name"),
+                        "lat": float(result.get("lat")),
+                        "lon": float(result.get("lon")),
+                        "osm_type": result.get("osm_type"),
+                        "osm_id": result.get("osm_id"),
+                        "place_id": result.get("place_id"),
+                        "type": result.get("type"),
+                        "class": result.get("class"),
+                        "address": result.get("address", {}),
+                        "boundingbox": result.get("boundingbox", [])
+                    })
+
+                logger.info(
+                    "Address search completed",
+                    query=query,
+                    result_count=len(addresses)
+                )
+
+                return addresses
+
+        except httpx.HTTPError as e:
+            logger.error(
+                "Nominatim API request failed",
+                query=query,
+                error=str(e)
+            )
+            return []
+        except Exception as e:
+            logger.error(
+                "Unexpected error in address search",
+                query=query,
+                error=str(e),
+                exc_info=True
+            )
+            return []
+
+    async def geocode_address(
+        self,
+        address: str,
+        country_code: str = "es"
+    ) -> Optional[Dict[str, Any]]:
+        """
+        Geocode an address to get coordinates.
+
+        Args:
+            address: Full address string
+            country_code: ISO country code
+
+        Returns:
+            Dictionary with lat, lon, display_name, address components or None
+        """
+        results = await self.search_address(address, country_code, limit=1)
+
+        if not results:
+            logger.warning("No geocoding results found", address=address)
+            return None
+
+        result = results[0]
+
+        logger.info(
+            "Address geocoded successfully",
+            address=address,
+            lat=result["lat"],
+            lon=result["lon"]
+        )
+
+        return result
+
+    async def reverse_geocode(
+        self,
+        latitude: float,
+        longitude: float
+    ) -> Optional[Dict[str, Any]]:
+        """
+        Reverse geocode coordinates to get address.
+
+        Args:
+            latitude: Latitude coordinate
+            longitude: Longitude coordinate
+
+        Returns:
+            Dictionary with address information or None
+        """
+        try:
+            # Rate limiting for public API
+            if self.is_public_api:
+                await sleep(self.min_request_interval)
+
+            async with httpx.AsyncClient(timeout=10.0) as client:
+                response = await client.get(
+                    f"{self.base_url}/reverse",
+                    params={
+                        "lat": latitude,
+                        "lon": longitude,
+                        "format": "json",
+                        "addressdetails": 1,
+                        "accept-language": "es"
+                    },
+                    headers=self.headers
+                )
+                response.raise_for_status()
+                result = response.json()
+
+                address_info = {
+                    "display_name": result.get("display_name"),
+                    "lat": float(result.get("lat")),
+                    "lon": float(result.get("lon")),
+                    "osm_type": result.get("osm_type"),
+                    "osm_id": result.get("osm_id"),
+                    "place_id": result.get("place_id"),
+                    "address": result.get("address", {}),
+                    "boundingbox": result.get("boundingbox", [])
+                }
+
+                logger.info(
+                    "Reverse geocoding completed",
+                    lat=latitude,
+                    lon=longitude,
+                    address=address_info["display_name"]
+                )
+
+                return address_info
+
+        except httpx.HTTPError as e:
+            logger.error(
+                "Nominatim reverse geocoding failed",
+                lat=latitude,
+                lon=longitude,
+                error=str(e)
+            )
+            return None
+        except Exception as e:
+            logger.error(
+                "Unexpected error in reverse geocoding",
+                lat=latitude,
+                lon=longitude,
+                error=str(e),
+                exc_info=True
+            )
+            return None
+
+    async def validate_coordinates(
+        self,
+        latitude: float,
+        longitude: float
+    ) -> bool:
+        """
+        Validate that coordinates point to a real location.
+
+        Args:
+            latitude: Latitude to validate
+            longitude: Longitude to validate
+
+        Returns:
+            True if coordinates are valid, False otherwise
+        """
+        if not (-90 <= latitude <= 90 and -180 <= longitude <= 180):
+            return False
+
+        result = await self.reverse_geocode(latitude, longitude)
+        return result is not None
+
+    async def health_check(self) -> bool:
+        """
+        Check if Nominatim service is accessible.
+
+        Returns:
+            True if service is healthy, False otherwise
+        """
+        try:
+            async with httpx.AsyncClient(timeout=5.0) as client:
+                response = await client.get(
+                    f"{self.base_url}/status",
+                    params={"format": "json"},
+                    headers=self.headers
+                )
+                return response.status_code == 200
+        except Exception as e:
+            logger.error(
+                "Nominatim health check failed",
+                error=str(e)
+            )
+            return False

services/external/app/services/poi_detection_service.py

@@ -0,0 +1,466 @@
+"""
+POI Detection Service
+
+Automated Point of Interest detection using Overpass API (OpenStreetMap).
+Detects nearby POIs around bakery locations and generates ML features
+for location-based demand forecasting.
+"""
+
+import overpy
+from typing import List, Dict, Any, Tuple, Optional
+from datetime import datetime, timezone, timedelta
+import asyncio
+import structlog
+import httpx
+from math import radians, sin, cos, sqrt, atan2
+import random
+
+from app.core.poi_config import (
+    POI_CATEGORIES,
+    OVERPASS_API_URL,
+    OVERPASS_TIMEOUT_SECONDS,
+    OVERPASS_MAX_RETRIES,
+    OVERPASS_RETRY_DELAY_SECONDS,
+    DISTANCE_BANDS
+)
+
+logger = structlog.get_logger()
+
+
+class POIDetectionService:
+    """
+    Automated POI detection using Overpass API (OpenStreetMap).
+
+    Detects points of interest near bakery locations and calculates
+    ML features for demand forecasting with location-specific context.
+    """
+
+    def __init__(self, overpass_url: str = OVERPASS_API_URL):
+        self.overpass_url = overpass_url
+        self.api = overpy.Overpass(url=overpass_url)
+        self.timeout = OVERPASS_TIMEOUT_SECONDS
+
+    async def detect_pois_for_bakery(
+        self,
+        latitude: float,
+        longitude: float,
+        tenant_id: str
+    ) -> Dict[str, Any]:
+        """
+        Detect all POIs around a bakery location.
+
+        Args:
+            latitude: Bakery latitude
+            longitude: Bakery longitude
+            tenant_id: Tenant identifier for logging
+
+        Returns:
+            Complete POI detection results with ML features
+        """
+        logger.info(
+            "Starting POI detection",
+            tenant_id=tenant_id,
+            location=(latitude, longitude)
+        )
+
+        poi_results = {}
+        detection_errors = []
+
+        # Query each POI category with inter-query delays
+        category_items = list(POI_CATEGORIES.items())
+        for idx, (category_key, category) in enumerate(category_items):
+            try:
+                pois = await self._query_pois_with_retry(
+                    latitude,
+                    longitude,
+                    category.osm_query,
+                    category.search_radius_m,
+                    category_key
+                )
+
+                # Calculate features for this category
+                features = self._calculate_poi_features(
+                    pois,
+                    (latitude, longitude),
+                    category
+                )
+
+                poi_results[category_key] = {
+                    "pois": pois,
+                    "features": features,
+                    "count": len(pois)
+                }
+
+                logger.info(
+                    f"Detected {category_key}",
+                    count=len(pois),
+                    proximity_score=features["proximity_score"]
+                )
+
+                # Add delay between categories to respect rate limits
+                # (except after the last category)
+                if idx < len(category_items) - 1:
+                    inter_query_delay = 2.0 + random.uniform(0.5, 1.5)
+                    await asyncio.sleep(inter_query_delay)
+
+            except Exception as e:
+                logger.error(
+                    f"Failed to detect {category_key}",
+                    error=str(e),
+                    tenant_id=tenant_id
+                )
+                detection_errors.append({
+                    "category": category_key,
+                    "error": str(e)
+                })
+                poi_results[category_key] = {
+                    "pois": [],
+                    "features": self._get_empty_features(),
+                    "count": 0,
+                    "error": str(e)
+                }
+
+                # Add a longer delay after an error before continuing
+                if idx < len(category_items) - 1:
+                    error_recovery_delay = 3.0 + random.uniform(1.0, 2.0)
+                    await asyncio.sleep(error_recovery_delay)
+
+        # Generate combined ML features
+        ml_features = self._generate_ml_features(poi_results)
+
+        # Generate summary
+        summary = self._generate_summary(poi_results)
+
+        detection_status = "completed" if not detection_errors else (
+            "partial" if len(detection_errors) < len(POI_CATEGORIES) else "failed"
+        )
+
+        return {
+            "tenant_id": tenant_id,
+            "location": {"latitude": latitude, "longitude": longitude},
+            "detection_timestamp": datetime.now(timezone.utc).isoformat(),
+            "detection_status": detection_status,
+            "detection_errors": detection_errors if detection_errors else None,
+            "poi_categories": poi_results,
+            "ml_features": ml_features,
+            "summary": summary
+        }
+
+    async def _query_pois_with_retry(
+        self,
+        latitude: float,
+        longitude: float,
+        osm_query: str,
+        radius_m: int,
+        category_key: str
+    ) -> List[Dict[str, Any]]:
+        """
+        Query Overpass API with exponential backoff retry logic.
+
+        Implements:
+        - Exponential backoff with jitter
+        - Extended delays for rate limiting errors
+        - Proper error type detection
+        """
+        last_error = None
+        base_delay = OVERPASS_RETRY_DELAY_SECONDS
+
+        for attempt in range(OVERPASS_MAX_RETRIES):
+            try:
+                return await self._query_pois(
+                    latitude, longitude, osm_query, radius_m
+                )
+            except Exception as e:
+                last_error = e
+                error_message = str(e).lower()
+
+                # Determine if this is a rate limiting error
+                is_rate_limit = any(phrase in error_message for phrase in [
+                    'too many requests',
+                    'rate limit',
+                    'server load too high',
+                    'quota exceeded',
+                    'retry later',
+                    '429',
+                    '503',
+                    '504'
+                ])
+
+                if attempt < OVERPASS_MAX_RETRIES - 1:
+                    # Calculate exponential backoff with jitter
+                    # For rate limiting: use longer delays (10-30 seconds)
+                    # For other errors: use standard backoff (2-8 seconds)
+                    if is_rate_limit:
+                        delay = base_delay * (3 ** attempt) + random.uniform(1, 5)
+                        delay = min(delay, 30)  # Cap at 30 seconds
+                    else:
+                        delay = base_delay * (2 ** attempt) + random.uniform(0.5, 1.5)
+                        delay = min(delay, 10)  # Cap at 10 seconds
+
+                    logger.warning(
+                        f"POI query retry {attempt + 1}/{OVERPASS_MAX_RETRIES}",
+                        category=category_key,
+                        error=str(e),
+                        is_rate_limit=is_rate_limit,
+                        retry_delay=f"{delay:.1f}s"
+                    )
+                    await asyncio.sleep(delay)
+                else:
+                    logger.error(
+                        "POI query failed after all retries",
+                        category=category_key,
+                        error=str(e),
+                        is_rate_limit=is_rate_limit
+                    )
+
+        raise last_error
+
+    async def _query_pois(
+        self,
+        latitude: float,
+        longitude: float,
+        osm_query: str,
+        radius_m: int
+    ) -> List[Dict[str, Any]]:
+        """
+        Query Overpass API for POIs in radius.
+
+        Raises:
+            Exception: With descriptive error message from Overpass API
+        """
+
+        # Build Overpass QL query
+        query = f"""
+        [out:json][timeout:{self.timeout}];
+        (
+          node{osm_query}(around:{radius_m},{latitude},{longitude});
+          way{osm_query}(around:{radius_m},{latitude},{longitude});
+        );
+        out center;
+        """
+
+        # Execute query (use asyncio thread pool for blocking overpy)
+        loop = asyncio.get_event_loop()
+        try:
+            result = await loop.run_in_executor(
+                None,
+                self.api.query,
+                query
+            )
+        except overpy.exception.OverpassTooManyRequests as e:
+            # Explicitly handle rate limiting
+            raise Exception("Too many requests - Overpass API rate limit exceeded") from e
+        except overpy.exception.OverpassGatewayTimeout as e:
+            # Query took too long
+            raise Exception("Gateway timeout - query too complex or server busy") from e
+        except overpy.exception.OverpassBadRequest as e:
+            # Query syntax error
+            raise Exception(f"Bad request - invalid query syntax: {str(e)}") from e
+        except Exception as e:
+            # Check if it's an HTTP error with status code
+            error_msg = str(e).lower()
+            if '429' in error_msg or 'too many' in error_msg:
+                raise Exception("Too many requests - rate limit exceeded") from e
+            elif '503' in error_msg or 'load too high' in error_msg:
+                raise Exception("Server load too high - Overpass API overloaded") from e
+            elif '504' in error_msg or 'timeout' in error_msg:
+                raise Exception("Gateway timeout - server busy") from e
+            else:
+                # Re-raise with original message
+                raise
+
+        # Parse results
+        pois = []
+
+        # Process nodes
+        for node in result.nodes:
+            pois.append({
+                "osm_id": str(node.id),
+                "type": "node",
+                "lat": float(node.lat),
+                "lon": float(node.lon),
+                "tags": dict(node.tags),
+                "name": node.tags.get("name", "Unnamed")
+            })
+
+        # Process ways (buildings, areas)
+        for way in result.ways:
+            # Get center point
+            if hasattr(way, 'center_lat') and way.center_lat:
+                lat, lon = float(way.center_lat), float(way.center_lon)
+            else:
+                # Calculate centroid from nodes
+                if way.nodes:
+                    lats = [float(node.lat) for node in way.nodes]
+                    lons = [float(node.lon) for node in way.nodes]
+                    lat = sum(lats) / len(lats)
+                    lon = sum(lons) / len(lons)
+                else:
+                    continue
+
+            pois.append({
+                "osm_id": str(way.id),
+                "type": "way",
+                "lat": lat,
+                "lon": lon,
+                "tags": dict(way.tags),
+                "name": way.tags.get("name", "Unnamed")
+            })
+
+        return pois
+
+    def _calculate_poi_features(
+        self,
+        pois: List[Dict[str, Any]],
+        bakery_location: Tuple[float, float],
+        category
+    ) -> Dict[str, float]:
+        """Calculate ML features for POI category"""
+
+        if not pois:
+            return self._get_empty_features()
+
+        # Calculate distances
+        distances = []
+        for poi in pois:
+            dist_km = self._haversine_distance(
+                bakery_location,
+                (poi["lat"], poi["lon"])
+            )
+            distances.append(dist_km * 1000)  # Convert to meters
+
+        # Feature Tier 1: Proximity Scores (PRIMARY)
+        proximity_score = sum(1.0 / (1.0 + d/1000) for d in distances)
+        weighted_proximity_score = proximity_score * category.weight
+
+        # Feature Tier 2: Distance Band Counts
+        count_0_100m = sum(1 for d in distances if d <= 100)
+        count_100_300m = sum(1 for d in distances if 100 < d <= 300)
+        count_300_500m = sum(1 for d in distances if 300 < d <= 500)
+        count_500_1000m = sum(1 for d in distances if 500 < d <= 1000)
+
+        # Feature Tier 3: Distance to Nearest
+        distance_to_nearest_m = min(distances) if distances else 9999.0
+
+        # Feature Tier 4: Binary Flags
+        has_within_100m = any(d <= 100 for d in distances)
+        has_within_300m = any(d <= 300 for d in distances)
+        has_within_500m = any(d <= 500 for d in distances)
+
+        return {
+            # Tier 1: Proximity scores (PRIMARY for ML)
+            "proximity_score": round(proximity_score, 4),
+            "weighted_proximity_score": round(weighted_proximity_score, 4),
+
+            # Tier 2: Distance bands
+            "count_0_100m": count_0_100m,
+            "count_100_300m": count_100_300m,
+            "count_300_500m": count_300_500m,
+            "count_500_1000m": count_500_1000m,
+            "total_count": len(pois),
+
+            # Tier 3: Distance to nearest
+            "distance_to_nearest_m": round(distance_to_nearest_m, 1),
+
+            # Tier 4: Binary flags
+            "has_within_100m": has_within_100m,
+            "has_within_300m": has_within_300m,
+            "has_within_500m": has_within_500m
+        }
+
+    def _generate_ml_features(self, poi_results: Dict[str, Any]) -> Dict[str, float]:
+        """
+        Generate flat feature dictionary for ML model ingestion.
+
+        These features will be added to Prophet/XGBoost as regressors.
+        """
+        ml_features = {}
+
+        for category_key, data in poi_results.items():
+            features = data.get("features", {})
+
+            # Flatten with category prefix
+            for feature_name, value in features.items():
+                ml_feature_name = f"poi_{category_key}_{feature_name}"
+                # Convert boolean to int for ML
+                if isinstance(value, bool):
+                    value = 1 if value else 0
+                ml_features[ml_feature_name] = value
+
+        return ml_features
+
+    def _get_empty_features(self) -> Dict[str, float]:
+        """Return zero features when no POIs found"""
+        return {
+            "proximity_score": 0.0,
+            "weighted_proximity_score": 0.0,
+            "count_0_100m": 0,
+            "count_100_300m": 0,
+            "count_300_500m": 0,
+            "count_500_1000m": 0,
+            "total_count": 0,
+            "distance_to_nearest_m": 9999.0,
+            "has_within_100m": False,
+            "has_within_300m": False,
+            "has_within_500m": False
+        }
+
+    def _haversine_distance(
+        self,
+        coord1: Tuple[float, float],
+        coord2: Tuple[float, float]
+    ) -> float:
+        """
+        Calculate distance between two coordinates in kilometers.
+
+        Uses Haversine formula for great-circle distance.
+        """
+        lat1, lon1 = coord1
+        lat2, lon2 = coord2
+
+        R = 6371  # Earth radius in km
+
+        dlat = radians(lat2 - lat1)
+        dlon = radians(lon2 - lon1)
+
+        a = (sin(dlat/2)**2 +
+             cos(radians(lat1)) * cos(radians(lat2)) * sin(dlon/2)**2)
+        c = 2 * atan2(sqrt(a), sqrt(1-a))
+
+        return R * c
+
+    def _generate_summary(self, poi_results: Dict[str, Any]) -> Dict[str, Any]:
+        """Generate human-readable summary"""
+        total_pois = sum(r["count"] for r in poi_results.values())
+        categories_with_pois = [
+            k for k, v in poi_results.items() if v["count"] > 0
+        ]
+        high_impact_categories = [
+            k for k, v in poi_results.items()
+            if v["features"]["proximity_score"] > 2.0
+        ]
+
+        return {
+            "total_pois_detected": total_pois,
+            "categories_with_pois": categories_with_pois,
+            "high_impact_categories": high_impact_categories,
+            "categories_count": len(categories_with_pois)
+        }
+
+    async def health_check(self) -> Dict[str, Any]:
+        """Check if Overpass API is accessible"""
+        try:
+            async with httpx.AsyncClient(timeout=5) as client:
+                response = await client.get(f"{self.overpass_url}/status")
+                is_healthy = response.status_code == 200
+                return {
+                    "healthy": is_healthy,
+                    "status_code": response.status_code,
+                    "url": self.overpass_url
+                }
+        except Exception as e:
+            return {
+                "healthy": False,
+                "error": str(e),
+                "url": self.overpass_url
+            }

services/external/app/services/poi_feature_selector.py

@@ -0,0 +1,184 @@
+"""
+POI Feature Selector
+
+Determines which POI features are relevant for ML model inclusion.
+Filters out low-signal features to prevent model noise and overfitting.
+"""
+
+from typing import Dict, List, Any
+import structlog
+
+from app.core.poi_config import RELEVANCE_THRESHOLDS
+
+logger = structlog.get_logger()
+
+
+class POIFeatureSelector:
+    """
+    Feature relevance engine for POI-based ML features.
+
+    Applies research-based thresholds to filter out irrelevant POI features
+    that would add noise to bakery-specific demand forecasting models.
+    """
+
+    def __init__(self, thresholds: Dict[str, Dict[str, float]] = None):
+        """
+        Initialize feature selector.
+
+        Args:
+            thresholds: Custom relevance thresholds (defaults to RELEVANCE_THRESHOLDS)
+        """
+        self.thresholds = thresholds or RELEVANCE_THRESHOLDS
+
+    def select_relevant_features(
+        self,
+        poi_detection_results: Dict[str, Any],
+        tenant_id: str = None
+    ) -> Dict[str, Any]:
+        """
+        Filter POI features based on relevance thresholds.
+
+        Only includes features for POI categories that pass relevance tests.
+        This prevents adding noise to ML models for bakeries where certain
+        POI categories are not significant.
+
+        Args:
+            poi_detection_results: Full POI detection results
+            tenant_id: Optional tenant ID for logging
+
+        Returns:
+            Dictionary with relevant features and detailed relevance report
+        """
+        relevant_features = {}
+        relevance_report = []
+        relevant_categories = []
+
+        for category_key, data in poi_detection_results.items():
+            features = data.get("features", {})
+            thresholds = self.thresholds.get(category_key, {})
+
+            if not thresholds:
+                logger.warning(
+                    f"No thresholds defined for category {category_key}",
+                    tenant_id=tenant_id
+                )
+                continue
+
+            # Check relevance criteria
+            is_relevant, rejection_reason = self._check_relevance(
+                features, thresholds, category_key
+            )
+
+            if is_relevant:
+                # Include features with category prefix
+                for feature_name, value in features.items():
+                    ml_feature_name = f"poi_{category_key}_{feature_name}"
+                    # Convert boolean to int for ML
+                    if isinstance(value, bool):
+                        value = 1 if value else 0
+                    relevant_features[ml_feature_name] = value
+
+                relevant_categories.append(category_key)
+                relevance_report.append({
+                    "category": category_key,
+                    "relevant": True,
+                    "reason": "Passes all relevance thresholds",
+                    "proximity_score": features.get("proximity_score", 0),
+                    "count": features.get("total_count", 0),
+                    "distance_to_nearest_m": features.get("distance_to_nearest_m", 9999)
+                })
+            else:
+                relevance_report.append({
+                    "category": category_key,
+                    "relevant": False,
+                    "reason": rejection_reason,
+                    "proximity_score": features.get("proximity_score", 0),
+                    "count": features.get("total_count", 0),
+                    "distance_to_nearest_m": features.get("distance_to_nearest_m", 9999)
+                })
+
+        logger.info(
+            "POI feature selection complete",
+            tenant_id=tenant_id,
+            total_categories=len(poi_detection_results),
+            relevant_categories=len(relevant_categories),
+            rejected_categories=len(poi_detection_results) - len(relevant_categories)
+        )
+
+        return {
+            "features": relevant_features,
+            "relevant_categories": relevant_categories,
+            "relevance_report": relevance_report,
+            "total_features": len(relevant_features),
+            "total_relevant_categories": len(relevant_categories)
+        }
+
+    def _check_relevance(
+        self,
+        features: Dict[str, Any],
+        thresholds: Dict[str, float],
+        category_key: str
+    ) -> tuple[bool, str]:
+        """
+        Check if POI category passes relevance thresholds.
+
+        Returns:
+            Tuple of (is_relevant, rejection_reason)
+        """
+        # Criterion 1: Proximity score
+        min_proximity = thresholds.get("min_proximity_score", 0)
+        actual_proximity = features.get("proximity_score", 0)
+        if actual_proximity < min_proximity:
+            return False, f"Proximity score too low ({actual_proximity:.2f} < {min_proximity})"
+
+        # Criterion 2: Distance to nearest
+        max_distance = thresholds.get("max_distance_to_nearest_m", 9999)
+        actual_distance = features.get("distance_to_nearest_m", 9999)
+        if actual_distance > max_distance:
+            return False, f"Nearest POI too far ({actual_distance:.0f}m > {max_distance}m)"
+
+        # Criterion 3: Count threshold
+        min_count = thresholds.get("min_count", 0)
+        actual_count = features.get("total_count", 0)
+        if actual_count < min_count:
+            return False, f"Count too low ({actual_count} < {min_count})"
+
+        return True, "Passes all thresholds"
+
+    def get_feature_importance_summary(
+        self,
+        poi_detection_results: Dict[str, Any]
+    ) -> List[Dict[str, Any]]:
+        """
+        Generate summary of feature importance for all categories.
+
+        Useful for understanding POI landscape around a bakery.
+        """
+        summary = []
+
+        for category_key, data in poi_detection_results.items():
+            features = data.get("features", {})
+            thresholds = self.thresholds.get(category_key, {})
+
+            is_relevant, reason = self._check_relevance(
+                features, thresholds, category_key
+            ) if thresholds else (False, "No thresholds defined")
+
+            summary.append({
+                "category": category_key,
+                "is_relevant": is_relevant,
+                "proximity_score": features.get("proximity_score", 0),
+                "weighted_score": features.get("weighted_proximity_score", 0),
+                "total_count": features.get("total_count", 0),
+                "distance_to_nearest_m": features.get("distance_to_nearest_m", 9999),
+                "has_within_100m": features.get("has_within_100m", False),
+                "rejection_reason": None if is_relevant else reason
+            })
+
+        # Sort by relevance and proximity score
+        summary.sort(
+            key=lambda x: (x["is_relevant"], x["proximity_score"]),
+            reverse=True
+        )
+
+        return summary

services/external/app/services/poi_refresh_service.py

@@ -0,0 +1,468 @@
+"""
+POI Refresh Service
+
+Manages periodic POI context refresh jobs.
+Detects changes in POI landscape and updates tenant POI contexts.
+"""
+
+import asyncio
+from datetime import datetime, timezone, timedelta
+from typing import Optional, Dict, Any, List
+from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy import select, and_, or_
+import structlog
+
+from app.models.poi_refresh_job import POIRefreshJob
+from app.models.poi_context import TenantPOIContext
+from app.services.poi_detection_service import POIDetectionService
+from app.core.database import database_manager
+
+logger = structlog.get_logger()
+
+
+class POIRefreshService:
+    """
+    POI Refresh Service
+
+    Manages background jobs for periodic POI context refresh.
+    Default refresh cycle: 180 days (6 months).
+    """
+
+    DEFAULT_REFRESH_INTERVAL_DAYS = 180
+    DEFAULT_MAX_ATTEMPTS = 3
+    STALE_THRESHOLD_DAYS = 180
+
+    def __init__(
+        self,
+        poi_detection_service: Optional[POIDetectionService] = None,
+        refresh_interval_days: int = DEFAULT_REFRESH_INTERVAL_DAYS
+    ):
+        """
+        Initialize POI refresh service.
+
+        Args:
+            poi_detection_service: POI detection service instance
+            refresh_interval_days: Days between POI refreshes (default: 180)
+        """
+        self.poi_detection_service = poi_detection_service or POIDetectionService()
+        self.refresh_interval_days = refresh_interval_days
+
+        logger.info(
+            "POI Refresh Service initialized",
+            refresh_interval_days=refresh_interval_days
+        )
+
+    async def schedule_refresh_job(
+        self,
+        tenant_id: str,
+        latitude: float,
+        longitude: float,
+        scheduled_at: Optional[datetime] = None,
+        session: Optional[AsyncSession] = None
+    ) -> POIRefreshJob:
+        """
+        Schedule a POI refresh job for a tenant.
+
+        Args:
+            tenant_id: Tenant UUID
+            latitude: Bakery latitude
+            longitude: Bakery longitude
+            scheduled_at: When to run the job (default: now + refresh_interval)
+            session: Database session
+
+        Returns:
+            Created POIRefreshJob
+        """
+        if scheduled_at is None:
+            scheduled_at = datetime.now(timezone.utc) + timedelta(
+                days=self.refresh_interval_days
+            )
+
+        async def _create_job(db_session: AsyncSession):
+            # Check if pending job already exists
+            result = await db_session.execute(
+                select(POIRefreshJob).where(
+                    and_(
+                        POIRefreshJob.tenant_id == tenant_id,
+                        POIRefreshJob.status.in_(["pending", "running"])
+                    )
+                )
+            )
+            existing_job = result.scalar_one_or_none()
+
+            if existing_job:
+                logger.info(
+                    "POI refresh job already scheduled",
+                    tenant_id=tenant_id,
+                    job_id=str(existing_job.id),
+                    scheduled_at=existing_job.scheduled_at
+                )
+                return existing_job
+
+            # Create new job
+            job = POIRefreshJob(
+                tenant_id=tenant_id,
+                latitude=latitude,
+                longitude=longitude,
+                scheduled_at=scheduled_at,
+                status="pending",
+                max_attempts=self.DEFAULT_MAX_ATTEMPTS
+            )
+
+            db_session.add(job)
+            await db_session.commit()
+            await db_session.refresh(job)
+
+            logger.info(
+                "POI refresh job scheduled",
+                tenant_id=tenant_id,
+                job_id=str(job.id),
+                scheduled_at=scheduled_at
+            )
+
+            return job
+
+        if session:
+            return await _create_job(session)
+        else:
+            async with database_manager.get_session() as db_session:
+                return await _create_job(db_session)
+
+    async def execute_refresh_job(
+        self,
+        job_id: str,
+        session: Optional[AsyncSession] = None
+    ) -> Dict[str, Any]:
+        """
+        Execute a POI refresh job.
+
+        Args:
+            job_id: Job UUID
+            session: Database session
+
+        Returns:
+            Execution result with status and details
+        """
+        async def _execute(db_session: AsyncSession):
+            # Load job
+            result = await db_session.execute(
+                select(POIRefreshJob).where(POIRefreshJob.id == job_id)
+            )
+            job = result.scalar_one_or_none()
+
+            if not job:
+                raise ValueError(f"Job not found: {job_id}")
+
+            if job.status == "running":
+                return {
+                    "status": "already_running",
+                    "job_id": str(job.id),
+                    "message": "Job is already running"
+                }
+
+            if job.status == "completed":
+                return {
+                    "status": "already_completed",
+                    "job_id": str(job.id),
+                    "message": "Job already completed"
+                }
+
+            if not job.can_retry:
+                return {
+                    "status": "max_attempts_reached",
+                    "job_id": str(job.id),
+                    "message": f"Max attempts ({job.max_attempts}) reached"
+                }
+
+            # Update job status
+            job.status = "running"
+            job.started_at = datetime.now(timezone.utc)
+            job.attempt_count += 1
+            await db_session.commit()
+
+            logger.info(
+                "Executing POI refresh job",
+                job_id=str(job.id),
+                tenant_id=str(job.tenant_id),
+                attempt=job.attempt_count
+            )
+
+            try:
+                # Get existing POI context
+                poi_result = await db_session.execute(
+                    select(TenantPOIContext).where(
+                        TenantPOIContext.tenant_id == job.tenant_id
+                    )
+                )
+                existing_context = poi_result.scalar_one_or_none()
+
+                # Perform POI detection
+                detection_result = await self.poi_detection_service.detect_pois_for_bakery(
+                    latitude=job.latitude,
+                    longitude=job.longitude,
+                    tenant_id=str(job.tenant_id),
+                    force_refresh=True
+                )
+
+                # Analyze changes
+                changes = self._analyze_changes(
+                    existing_context.poi_detection_results if existing_context else {},
+                    detection_result
+                )
+
+                # Update job with results
+                job.status = "completed"
+                job.completed_at = datetime.now(timezone.utc)
+                job.pois_detected = sum(
+                    data.get("count", 0)
+                    for data in detection_result.values()
+                )
+                job.changes_detected = changes["has_significant_changes"]
+                job.change_summary = changes
+
+                # Schedule next refresh
+                job.next_scheduled_at = datetime.now(timezone.utc) + timedelta(
+                    days=self.refresh_interval_days
+                )
+
+                await db_session.commit()
+
+                logger.info(
+                    "POI refresh job completed",
+                    job_id=str(job.id),
+                    tenant_id=str(job.tenant_id),
+                    pois_detected=job.pois_detected,
+                    changes_detected=job.changes_detected,
+                    duration_seconds=job.duration_seconds
+                )
+
+                # Schedule next job
+                await self.schedule_refresh_job(
+                    tenant_id=str(job.tenant_id),
+                    latitude=job.latitude,
+                    longitude=job.longitude,
+                    scheduled_at=job.next_scheduled_at,
+                    session=db_session
+                )
+
+                return {
+                    "status": "success",
+                    "job_id": str(job.id),
+                    "pois_detected": job.pois_detected,
+                    "changes_detected": job.changes_detected,
+                    "change_summary": changes,
+                    "duration_seconds": job.duration_seconds,
+                    "next_scheduled_at": job.next_scheduled_at.isoformat()
+                }
+
+            except Exception as e:
+                # Job failed
+                job.status = "failed"
+                job.completed_at = datetime.now(timezone.utc)
+                job.error_message = str(e)
+                job.error_details = {
+                    "error_type": type(e).__name__,
+                    "error_message": str(e),
+                    "attempt": job.attempt_count
+                }
+
+                # Schedule retry if attempts remaining
+                if job.can_retry:
+                    job.next_scheduled_at = datetime.now(timezone.utc) + timedelta(hours=1)
+                    logger.warning(
+                        "POI refresh job failed, will retry",
+                        job_id=str(job.id),
+                        tenant_id=str(job.tenant_id),
+                        attempt=job.attempt_count,
+                        max_attempts=job.max_attempts,
+                        error=str(e)
+                    )
+                else:
+                    logger.error(
+                        "POI refresh job failed permanently",
+                        job_id=str(job.id),
+                        tenant_id=str(job.tenant_id),
+                        attempt=job.attempt_count,
+                        error=str(e),
+                        exc_info=True
+                    )
+
+                await db_session.commit()
+
+                return {
+                    "status": "failed",
+                    "job_id": str(job.id),
+                    "error": str(e),
+                    "attempt": job.attempt_count,
+                    "can_retry": job.can_retry
+                }
+
+        if session:
+            return await _execute(session)
+        else:
+            async with database_manager.get_session() as db_session:
+                return await _execute(db_session)
+
+    def _analyze_changes(
+        self,
+        old_results: Dict[str, Any],
+        new_results: Dict[str, Any]
+    ) -> Dict[str, Any]:
+        """
+        Analyze changes between old and new POI detection results.
+
+        Args:
+            old_results: Previous POI detection results
+            new_results: New POI detection results
+
+        Returns:
+            Change analysis with significance flag
+        """
+        changes = {
+            "has_significant_changes": False,
+            "category_changes": {},
+            "total_poi_change": 0,
+            "new_categories": [],
+            "removed_categories": []
+        }
+
+        old_categories = set(old_results.keys())
+        new_categories = set(new_results.keys())
+
+        # New categories
+        changes["new_categories"] = list(new_categories - old_categories)
+
+        # Removed categories
+        changes["removed_categories"] = list(old_categories - new_categories)
+
+        # Analyze changes per category
+        for category in new_categories:
+            old_count = old_results.get(category, {}).get("count", 0)
+            new_count = new_results.get(category, {}).get("count", 0)
+            change = new_count - old_count
+
+            if abs(change) > 0:
+                changes["category_changes"][category] = {
+                    "old_count": old_count,
+                    "new_count": new_count,
+                    "change": change,
+                    "change_percent": (change / old_count * 100) if old_count > 0 else 100
+                }
+
+            changes["total_poi_change"] += abs(change)
+
+        # Determine if changes are significant
+        # Significant if: 10+ POIs changed OR 20%+ change OR new/removed categories
+        total_old_pois = sum(data.get("count", 0) for data in old_results.values())
+        if total_old_pois > 0:
+            change_percent = (changes["total_poi_change"] / total_old_pois) * 100
+            changes["total_change_percent"] = change_percent
+
+            changes["has_significant_changes"] = (
+                changes["total_poi_change"] >= 10
+                or change_percent >= 20
+                or len(changes["new_categories"]) > 0
+                or len(changes["removed_categories"]) > 0
+            )
+        else:
+            changes["has_significant_changes"] = changes["total_poi_change"] > 0
+
+        return changes
+
+    async def get_pending_jobs(
+        self,
+        limit: int = 100,
+        session: Optional[AsyncSession] = None
+    ) -> List[POIRefreshJob]:
+        """
+        Get pending jobs that are due for execution.
+
+        Args:
+            limit: Maximum number of jobs to return
+            session: Database session
+
+        Returns:
+            List of pending jobs
+        """
+        async def _get_jobs(db_session: AsyncSession):
+            result = await db_session.execute(
+                select(POIRefreshJob)
+                .where(
+                    and_(
+                        POIRefreshJob.status == "pending",
+                        POIRefreshJob.scheduled_at <= datetime.now(timezone.utc)
+                    )
+                )
+                .order_by(POIRefreshJob.scheduled_at)
+                .limit(limit)
+            )
+            return result.scalars().all()
+
+        if session:
+            return await _get_jobs(session)
+        else:
+            async with database_manager.get_session() as db_session:
+                return await _get_jobs(db_session)
+
+    async def process_pending_jobs(
+        self,
+        max_concurrent: int = 5,
+        session: Optional[AsyncSession] = None
+    ) -> Dict[str, Any]:
+        """
+        Process all pending jobs concurrently.
+
+        Args:
+            max_concurrent: Maximum concurrent job executions
+            session: Database session
+
+        Returns:
+            Processing summary
+        """
+        pending_jobs = await self.get_pending_jobs(session=session)
+
+        if not pending_jobs:
+            logger.info("No pending POI refresh jobs")
+            return {
+                "total_jobs": 0,
+                "successful": 0,
+                "failed": 0,
+                "results": []
+            }
+
+        logger.info(
+            "Processing pending POI refresh jobs",
+            count=len(pending_jobs),
+            max_concurrent=max_concurrent
+        )
+
+        # Process jobs with concurrency limit
+        semaphore = asyncio.Semaphore(max_concurrent)
+
+        async def process_job(job: POIRefreshJob):
+            async with semaphore:
+                return await self.execute_refresh_job(str(job.id))
+
+        results = await asyncio.gather(
+            *[process_job(job) for job in pending_jobs],
+            return_exceptions=True
+        )
+
+        # Summarize results
+        successful = sum(1 for r in results if isinstance(r, dict) and r.get("status") == "success")
+        failed = sum(1 for r in results if isinstance(r, dict) and r.get("status") == "failed")
+        errors = sum(1 for r in results if isinstance(r, Exception))
+
+        summary = {
+            "total_jobs": len(pending_jobs),
+            "successful": successful,
+            "failed": failed + errors,
+            "results": [r if not isinstance(r, Exception) else {"status": "error", "error": str(r)} for r in results]
+        }
+
+        logger.info(
+            "POI refresh jobs processing completed",
+            **summary
+        )
+
+        return summary

services/external/app/services/poi_scheduler.py

@@ -0,0 +1,187 @@
+"""
+POI Refresh Scheduler
+
+Background scheduler for periodic POI context refresh.
+Runs every hour to check for and execute pending POI refresh jobs.
+"""
+
+import asyncio
+from typing import Optional
+from datetime import datetime, timezone
+import structlog
+
+from app.services.poi_refresh_service import POIRefreshService
+
+logger = structlog.get_logger()
+
+
+class POIRefreshScheduler:
+    """
+    POI Refresh Scheduler
+
+    Background task that periodically checks for and executes
+    pending POI refresh jobs.
+    """
+
+    def __init__(
+        self,
+        poi_refresh_service: Optional[POIRefreshService] = None,
+        check_interval_seconds: int = 3600,  # 1 hour
+        max_concurrent_jobs: int = 5
+    ):
+        """
+        Initialize POI refresh scheduler.
+
+        Args:
+            poi_refresh_service: POI refresh service instance
+            check_interval_seconds: Seconds between checks (default: 3600 = 1 hour)
+            max_concurrent_jobs: Max concurrent job executions (default: 5)
+        """
+        self.poi_refresh_service = poi_refresh_service or POIRefreshService()
+        self.check_interval_seconds = check_interval_seconds
+        self.max_concurrent_jobs = max_concurrent_jobs
+
+        self._task: Optional[asyncio.Task] = None
+        self._running = False
+
+        logger.info(
+            "POI Refresh Scheduler initialized",
+            check_interval_seconds=check_interval_seconds,
+            max_concurrent_jobs=max_concurrent_jobs
+        )
+
+    async def start(self):
+        """Start the scheduler background task"""
+        if self._running:
+            logger.warning("POI Refresh Scheduler already running")
+            return
+
+        self._running = True
+        self._task = asyncio.create_task(self._run_scheduler())
+
+        logger.info("POI Refresh Scheduler started")
+
+    async def stop(self):
+        """Stop the scheduler background task"""
+        if not self._running:
+            return
+
+        self._running = False
+
+        if self._task:
+            self._task.cancel()
+            try:
+                await self._task
+            except asyncio.CancelledError:
+                pass
+
+        logger.info("POI Refresh Scheduler stopped")
+
+    async def _run_scheduler(self):
+        """Main scheduler loop"""
+        logger.info("POI Refresh Scheduler loop started")
+
+        while self._running:
+            try:
+                await self._process_cycle()
+            except Exception as e:
+                logger.error(
+                    "POI refresh scheduler cycle failed",
+                    error=str(e),
+                    exc_info=True
+                )
+
+            # Wait for next cycle
+            try:
+                await asyncio.sleep(self.check_interval_seconds)
+            except asyncio.CancelledError:
+                break
+
+        logger.info("POI Refresh Scheduler loop ended")
+
+    async def _process_cycle(self):
+        """Process one scheduler cycle"""
+        cycle_start = datetime.now(timezone.utc)
+
+        logger.debug(
+            "POI refresh scheduler cycle started",
+            timestamp=cycle_start.isoformat()
+        )
+
+        # Process pending jobs
+        result = await self.poi_refresh_service.process_pending_jobs(
+            max_concurrent=self.max_concurrent_jobs
+        )
+
+        cycle_end = datetime.now(timezone.utc)
+        cycle_duration = (cycle_end - cycle_start).total_seconds()
+
+        if result["total_jobs"] > 0:
+            logger.info(
+                "POI refresh scheduler cycle completed",
+                total_jobs=result["total_jobs"],
+                successful=result["successful"],
+                failed=result["failed"],
+                cycle_duration_seconds=cycle_duration
+            )
+        else:
+            logger.debug(
+                "POI refresh scheduler cycle completed (no jobs)",
+                cycle_duration_seconds=cycle_duration
+            )
+
+    async def trigger_immediate_check(self):
+        """Trigger an immediate check for pending jobs (bypasses schedule)"""
+        logger.info("POI refresh scheduler immediate check triggered")
+
+        try:
+            result = await self.poi_refresh_service.process_pending_jobs(
+                max_concurrent=self.max_concurrent_jobs
+            )
+
+            logger.info(
+                "POI refresh scheduler immediate check completed",
+                total_jobs=result["total_jobs"],
+                successful=result["successful"],
+                failed=result["failed"]
+            )
+
+            return result
+        except Exception as e:
+            logger.error(
+                "POI refresh scheduler immediate check failed",
+                error=str(e),
+                exc_info=True
+            )
+            raise
+
+    @property
+    def is_running(self) -> bool:
+        """Check if scheduler is running"""
+        return self._running
+
+
+# Global scheduler instance
+_scheduler_instance: Optional[POIRefreshScheduler] = None
+
+
+def get_scheduler() -> POIRefreshScheduler:
+    """Get global scheduler instance (singleton)"""
+    global _scheduler_instance
+
+    if _scheduler_instance is None:
+        _scheduler_instance = POIRefreshScheduler()
+
+    return _scheduler_instance
+
+
+async def start_scheduler():
+    """Start global POI refresh scheduler"""
+    scheduler = get_scheduler()
+    await scheduler.start()
+
+
+async def stop_scheduler():
+    """Stop global POI refresh scheduler"""
+    scheduler = get_scheduler()
+    await scheduler.stop()

services/external/migrations/versions/20251015_1230_b97bab14ac47_initial_schema_20251015_1230.py

@@ -1,314 +0,0 @@
-"""initial_schema_20251015_1230
-
-Revision ID: b97bab14ac47
-Revises: 
-Create Date: 2025-10-15 12:30:54.963197+02:00
-
-"""
-from typing import Sequence, Union
-
-from alembic import op
-import sqlalchemy as sa
-from sqlalchemy.dialects import postgresql
-
-# revision identifiers, used by Alembic.
-revision: str = 'b97bab14ac47'
-down_revision: Union[str, None] = None
-branch_labels: Union[str, Sequence[str], None] = None
-depends_on: Union[str, Sequence[str], None] = None
-
-
-def upgrade() -> None:
-    # ### commands auto generated by Alembic - please adjust! ###
-    op.create_table('audit_logs',
-    sa.Column('id', sa.UUID(), nullable=False),
-    sa.Column('tenant_id', sa.UUID(), nullable=False),
-    sa.Column('user_id', sa.UUID(), nullable=False),
-    sa.Column('action', sa.String(length=100), nullable=False),
-    sa.Column('resource_type', sa.String(length=100), nullable=False),
-    sa.Column('resource_id', sa.String(length=255), nullable=True),
-    sa.Column('severity', sa.String(length=20), nullable=False),
-    sa.Column('service_name', sa.String(length=100), nullable=False),
-    sa.Column('description', sa.Text(), nullable=True),
-    sa.Column('changes', postgresql.JSON(astext_type=sa.Text()), nullable=True),
-    sa.Column('audit_metadata', postgresql.JSON(astext_type=sa.Text()), nullable=True),
-    sa.Column('ip_address', sa.String(length=45), nullable=True),
-    sa.Column('user_agent', sa.Text(), nullable=True),
-    sa.Column('endpoint', sa.String(length=255), nullable=True),
-    sa.Column('method', sa.String(length=10), nullable=True),
-    sa.Column('created_at', sa.DateTime(timezone=True), nullable=False),
-    sa.PrimaryKeyConstraint('id')
-    )
-    op.create_index('idx_audit_resource_type_action', 'audit_logs', ['resource_type', 'action'], unique=False)
-    op.create_index('idx_audit_service_created', 'audit_logs', ['service_name', 'created_at'], unique=False)
-    op.create_index('idx_audit_severity_created', 'audit_logs', ['severity', 'created_at'], unique=False)
-    op.create_index('idx_audit_tenant_created', 'audit_logs', ['tenant_id', 'created_at'], unique=False)
-    op.create_index('idx_audit_user_created', 'audit_logs', ['user_id', 'created_at'], unique=False)
-    op.create_index(op.f('ix_audit_logs_action'), 'audit_logs', ['action'], unique=False)
-    op.create_index(op.f('ix_audit_logs_created_at'), 'audit_logs', ['created_at'], unique=False)
-    op.create_index(op.f('ix_audit_logs_resource_id'), 'audit_logs', ['resource_id'], unique=False)
-    op.create_index(op.f('ix_audit_logs_resource_type'), 'audit_logs', ['resource_type'], unique=False)
-    op.create_index(op.f('ix_audit_logs_service_name'), 'audit_logs', ['service_name'], unique=False)
-    op.create_index(op.f('ix_audit_logs_severity'), 'audit_logs', ['severity'], unique=False)
-    op.create_index(op.f('ix_audit_logs_tenant_id'), 'audit_logs', ['tenant_id'], unique=False)
-    op.create_index(op.f('ix_audit_logs_user_id'), 'audit_logs', ['user_id'], unique=False)
-    op.create_table('city_traffic_data',
-    sa.Column('id', sa.UUID(), nullable=False),
-    sa.Column('city_id', sa.String(length=50), nullable=False),
-    sa.Column('date', sa.DateTime(timezone=True), nullable=False),
-    sa.Column('traffic_volume', sa.Integer(), nullable=True),
-    sa.Column('pedestrian_count', sa.Integer(), nullable=True),
-    sa.Column('congestion_level', sa.String(length=20), nullable=True),
-    sa.Column('average_speed', sa.Float(), nullable=True),
-    sa.Column('source', sa.String(length=50), nullable=False),
-    sa.Column('raw_data', postgresql.JSONB(astext_type=sa.Text()), nullable=True),
-    sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
-    sa.PrimaryKeyConstraint('id')
-    )
-    op.create_index('idx_city_traffic_lookup', 'city_traffic_data', ['city_id', 'date'], unique=False)
-    op.create_index(op.f('ix_city_traffic_data_city_id'), 'city_traffic_data', ['city_id'], unique=False)
-    op.create_index(op.f('ix_city_traffic_data_date'), 'city_traffic_data', ['date'], unique=False)
-    op.create_table('city_weather_data',
-    sa.Column('id', sa.UUID(), nullable=False),
-    sa.Column('city_id', sa.String(length=50), nullable=False),
-    sa.Column('date', sa.DateTime(timezone=True), nullable=False),
-    sa.Column('temperature', sa.Float(), nullable=True),
-    sa.Column('precipitation', sa.Float(), nullable=True),
-    sa.Column('humidity', sa.Float(), nullable=True),
-    sa.Column('wind_speed', sa.Float(), nullable=True),
-    sa.Column('pressure', sa.Float(), nullable=True),
-    sa.Column('description', sa.String(length=200), nullable=True),
-    sa.Column('source', sa.String(length=50), nullable=False),
-    sa.Column('raw_data', postgresql.JSONB(astext_type=sa.Text()), nullable=True),
-    sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
-    sa.PrimaryKeyConstraint('id')
-    )
-    op.create_index('idx_city_weather_lookup', 'city_weather_data', ['city_id', 'date'], unique=False)
-    op.create_index(op.f('ix_city_weather_data_city_id'), 'city_weather_data', ['city_id'], unique=False)
-    op.create_index(op.f('ix_city_weather_data_date'), 'city_weather_data', ['date'], unique=False)
-    op.create_table('traffic_background_jobs',
-    sa.Column('id', sa.UUID(), nullable=False),
-    sa.Column('job_type', sa.String(length=50), nullable=False),
-    sa.Column('city', sa.String(length=50), nullable=False),
-    sa.Column('location_pattern', sa.String(length=200), nullable=True),
-    sa.Column('scheduled_at', sa.DateTime(timezone=True), nullable=False),
-    sa.Column('started_at', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('completed_at', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('status', sa.String(length=20), nullable=False),
-    sa.Column('progress_percentage', sa.Float(), nullable=True),
-    sa.Column('records_processed', sa.Integer(), nullable=True),
-    sa.Column('records_stored', sa.Integer(), nullable=True),
-    sa.Column('data_start_date', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('data_end_date', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('success_count', sa.Integer(), nullable=True),
-    sa.Column('error_count', sa.Integer(), nullable=True),
-    sa.Column('error_message', sa.Text(), nullable=True),
-    sa.Column('job_metadata', sa.JSON(), nullable=True),
-    sa.Column('tenant_id', sa.UUID(), nullable=True),
-    sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
-    sa.PrimaryKeyConstraint('id')
-    )
-    op.create_index('idx_jobs_city_status', 'traffic_background_jobs', ['city', 'status', 'scheduled_at'], unique=False)
-    op.create_index('idx_jobs_completed', 'traffic_background_jobs', ['status', 'completed_at'], unique=False)
-    op.create_index('idx_jobs_tenant_status', 'traffic_background_jobs', ['tenant_id', 'status', 'scheduled_at'], unique=False)
-    op.create_index('idx_jobs_type_city', 'traffic_background_jobs', ['job_type', 'city', 'scheduled_at'], unique=False)
-    op.create_index(op.f('ix_traffic_background_jobs_city'), 'traffic_background_jobs', ['city'], unique=False)
-    op.create_index(op.f('ix_traffic_background_jobs_tenant_id'), 'traffic_background_jobs', ['tenant_id'], unique=False)
-    op.create_table('traffic_data',
-    sa.Column('id', sa.UUID(), nullable=False),
-    sa.Column('location_id', sa.String(length=100), nullable=False),
-    sa.Column('city', sa.String(length=50), nullable=False),
-    sa.Column('date', sa.DateTime(timezone=True), nullable=False),
-    sa.Column('traffic_volume', sa.Integer(), nullable=True),
-    sa.Column('congestion_level', sa.String(length=20), nullable=True),
-    sa.Column('average_speed', sa.Float(), nullable=True),
-    sa.Column('occupation_percentage', sa.Float(), nullable=True),
-    sa.Column('load_percentage', sa.Float(), nullable=True),
-    sa.Column('pedestrian_count', sa.Integer(), nullable=True),
-    sa.Column('measurement_point_id', sa.String(length=100), nullable=True),
-    sa.Column('measurement_point_name', sa.String(length=500), nullable=True),
-    sa.Column('measurement_point_type', sa.String(length=50), nullable=True),
-    sa.Column('latitude', sa.Float(), nullable=True),
-    sa.Column('longitude', sa.Float(), nullable=True),
-    sa.Column('district', sa.String(length=100), nullable=True),
-    sa.Column('zone', sa.String(length=100), nullable=True),
-    sa.Column('source', sa.String(length=50), nullable=False),
-    sa.Column('data_quality_score', sa.Float(), nullable=True),
-    sa.Column('is_synthetic', sa.Boolean(), nullable=True),
-    sa.Column('has_pedestrian_inference', sa.Boolean(), nullable=True),
-    sa.Column('city_specific_data', sa.JSON(), nullable=True),
-    sa.Column('raw_data', sa.Text(), nullable=True),
-    sa.Column('tenant_id', sa.UUID(), nullable=True),
-    sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
-    sa.PrimaryKeyConstraint('id')
-    )
-    op.create_index('idx_traffic_city_date', 'traffic_data', ['city', 'date'], unique=False)
-    op.create_index('idx_traffic_city_location', 'traffic_data', ['city', 'location_id'], unique=False)
-    op.create_index('idx_traffic_district_date', 'traffic_data', ['city', 'district', 'date'], unique=False)
-    op.create_index('idx_traffic_location_date', 'traffic_data', ['location_id', 'date'], unique=False)
-    op.create_index('idx_traffic_measurement_point', 'traffic_data', ['city', 'measurement_point_id'], unique=False)
-    op.create_index('idx_traffic_quality', 'traffic_data', ['city', 'data_quality_score', 'date'], unique=False)
-    op.create_index('idx_traffic_tenant_date', 'traffic_data', ['tenant_id', 'date'], unique=False)
-    op.create_index('idx_traffic_training', 'traffic_data', ['tenant_id', 'city', 'date', 'is_synthetic'], unique=False)
-    op.create_index(op.f('ix_traffic_data_city'), 'traffic_data', ['city'], unique=False)
-    op.create_index(op.f('ix_traffic_data_date'), 'traffic_data', ['date'], unique=False)
-    op.create_index(op.f('ix_traffic_data_location_id'), 'traffic_data', ['location_id'], unique=False)
-    op.create_index(op.f('ix_traffic_data_measurement_point_id'), 'traffic_data', ['measurement_point_id'], unique=False)
-    op.create_index(op.f('ix_traffic_data_tenant_id'), 'traffic_data', ['tenant_id'], unique=False)
-    op.create_table('traffic_measurement_points',
-    sa.Column('id', sa.UUID(), nullable=False),
-    sa.Column('city', sa.String(length=50), nullable=False),
-    sa.Column('measurement_point_id', sa.String(length=100), nullable=False),
-    sa.Column('name', sa.String(length=500), nullable=True),
-    sa.Column('description', sa.Text(), nullable=True),
-    sa.Column('latitude', sa.Float(), nullable=False),
-    sa.Column('longitude', sa.Float(), nullable=False),
-    sa.Column('district', sa.String(length=100), nullable=True),
-    sa.Column('zone', sa.String(length=100), nullable=True),
-    sa.Column('road_type', sa.String(length=50), nullable=True),
-    sa.Column('measurement_type', sa.String(length=50), nullable=True),
-    sa.Column('point_category', sa.String(length=50), nullable=True),
-    sa.Column('is_active', sa.Boolean(), nullable=True),
-    sa.Column('installation_date', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('last_data_received', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('data_quality_rating', sa.Float(), nullable=True),
-    sa.Column('city_specific_metadata', sa.JSON(), nullable=True),
-    sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
-    sa.PrimaryKeyConstraint('id')
-    )
-    op.create_index('idx_points_active', 'traffic_measurement_points', ['city', 'is_active', 'last_data_received'], unique=False)
-    op.create_index('idx_points_city_location', 'traffic_measurement_points', ['city', 'latitude', 'longitude'], unique=False)
-    op.create_index('idx_points_district', 'traffic_measurement_points', ['city', 'district'], unique=False)
-    op.create_index('idx_points_road_type', 'traffic_measurement_points', ['city', 'road_type'], unique=False)
-    op.create_index('idx_unique_city_point', 'traffic_measurement_points', ['city', 'measurement_point_id'], unique=True)
-    op.create_index(op.f('ix_traffic_measurement_points_city'), 'traffic_measurement_points', ['city'], unique=False)
-    op.create_index(op.f('ix_traffic_measurement_points_measurement_point_id'), 'traffic_measurement_points', ['measurement_point_id'], unique=False)
-    op.create_table('weather_data',
-    sa.Column('id', sa.UUID(), nullable=False),
-    sa.Column('location_id', sa.String(length=100), nullable=False),
-    sa.Column('city', sa.String(length=50), nullable=False),
-    sa.Column('station_name', sa.String(length=200), nullable=True),
-    sa.Column('latitude', sa.Float(), nullable=True),
-    sa.Column('longitude', sa.Float(), nullable=True),
-    sa.Column('date', sa.DateTime(timezone=True), nullable=False),
-    sa.Column('forecast_date', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('temperature', sa.Float(), nullable=True),
-    sa.Column('temperature_min', sa.Float(), nullable=True),
-    sa.Column('temperature_max', sa.Float(), nullable=True),
-    sa.Column('feels_like', sa.Float(), nullable=True),
-    sa.Column('precipitation', sa.Float(), nullable=True),
-    sa.Column('precipitation_probability', sa.Float(), nullable=True),
-    sa.Column('humidity', sa.Float(), nullable=True),
-    sa.Column('wind_speed', sa.Float(), nullable=True),
-    sa.Column('wind_direction', sa.Float(), nullable=True),
-    sa.Column('wind_gust', sa.Float(), nullable=True),
-    sa.Column('pressure', sa.Float(), nullable=True),
-    sa.Column('visibility', sa.Float(), nullable=True),
-    sa.Column('uv_index', sa.Float(), nullable=True),
-    sa.Column('cloud_cover', sa.Float(), nullable=True),
-    sa.Column('condition', sa.String(length=100), nullable=True),
-    sa.Column('description', sa.String(length=200), nullable=True),
-    sa.Column('weather_code', sa.String(length=20), nullable=True),
-    sa.Column('source', sa.String(length=50), nullable=False),
-    sa.Column('data_type', sa.String(length=20), nullable=False),
-    sa.Column('is_forecast', sa.Boolean(), nullable=True),
-    sa.Column('data_quality_score', sa.Float(), nullable=True),
-    sa.Column('raw_data', postgresql.JSON(astext_type=sa.Text()), nullable=True),
-    sa.Column('processed_data', postgresql.JSON(astext_type=sa.Text()), nullable=True),
-    sa.Column('tenant_id', sa.UUID(), nullable=True),
-    sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
-    sa.PrimaryKeyConstraint('id')
-    )
-    op.create_index('idx_weather_location_date', 'weather_data', ['location_id', 'date'], unique=False)
-    op.create_index(op.f('ix_weather_data_date'), 'weather_data', ['date'], unique=False)
-    op.create_index(op.f('ix_weather_data_location_id'), 'weather_data', ['location_id'], unique=False)
-    op.create_index(op.f('ix_weather_data_tenant_id'), 'weather_data', ['tenant_id'], unique=False)
-    op.create_table('weather_forecasts',
-    sa.Column('id', sa.UUID(), nullable=False),
-    sa.Column('location_id', sa.String(length=100), nullable=False),
-    sa.Column('forecast_date', sa.DateTime(timezone=True), nullable=False),
-    sa.Column('generated_at', sa.DateTime(timezone=True), nullable=False),
-    sa.Column('temperature', sa.Float(), nullable=True),
-    sa.Column('precipitation', sa.Float(), nullable=True),
-    sa.Column('humidity', sa.Float(), nullable=True),
-    sa.Column('wind_speed', sa.Float(), nullable=True),
-    sa.Column('description', sa.String(length=200), nullable=True),
-    sa.Column('source', sa.String(length=50), nullable=False),
-    sa.Column('raw_data', sa.Text(), nullable=True),
-    sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
-    sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
-    sa.PrimaryKeyConstraint('id')
-    )
-    op.create_index('idx_forecast_location_date', 'weather_forecasts', ['location_id', 'forecast_date'], unique=False)
-    op.create_index(op.f('ix_weather_forecasts_location_id'), 'weather_forecasts', ['location_id'], unique=False)
-    # ### end Alembic commands ###
-
-
-def downgrade() -> None:
-    # ### commands auto generated by Alembic - please adjust! ###
-    op.drop_index(op.f('ix_weather_forecasts_location_id'), table_name='weather_forecasts')
-    op.drop_index('idx_forecast_location_date', table_name='weather_forecasts')
-    op.drop_table('weather_forecasts')
-    op.drop_index(op.f('ix_weather_data_tenant_id'), table_name='weather_data')
-    op.drop_index(op.f('ix_weather_data_location_id'), table_name='weather_data')
-    op.drop_index(op.f('ix_weather_data_date'), table_name='weather_data')
-    op.drop_index('idx_weather_location_date', table_name='weather_data')
-    op.drop_table('weather_data')
-    op.drop_index(op.f('ix_traffic_measurement_points_measurement_point_id'), table_name='traffic_measurement_points')
-    op.drop_index(op.f('ix_traffic_measurement_points_city'), table_name='traffic_measurement_points')
-    op.drop_index('idx_unique_city_point', table_name='traffic_measurement_points')
-    op.drop_index('idx_points_road_type', table_name='traffic_measurement_points')
-    op.drop_index('idx_points_district', table_name='traffic_measurement_points')
-    op.drop_index('idx_points_city_location', table_name='traffic_measurement_points')
-    op.drop_index('idx_points_active', table_name='traffic_measurement_points')
-    op.drop_table('traffic_measurement_points')
-    op.drop_index(op.f('ix_traffic_data_tenant_id'), table_name='traffic_data')
-    op.drop_index(op.f('ix_traffic_data_measurement_point_id'), table_name='traffic_data')
-    op.drop_index(op.f('ix_traffic_data_location_id'), table_name='traffic_data')
-    op.drop_index(op.f('ix_traffic_data_date'), table_name='traffic_data')
-    op.drop_index(op.f('ix_traffic_data_city'), table_name='traffic_data')
-    op.drop_index('idx_traffic_training', table_name='traffic_data')
-    op.drop_index('idx_traffic_tenant_date', table_name='traffic_data')
-    op.drop_index('idx_traffic_quality', table_name='traffic_data')
-    op.drop_index('idx_traffic_measurement_point', table_name='traffic_data')
-    op.drop_index('idx_traffic_location_date', table_name='traffic_data')
-    op.drop_index('idx_traffic_district_date', table_name='traffic_data')
-    op.drop_index('idx_traffic_city_location', table_name='traffic_data')
-    op.drop_index('idx_traffic_city_date', table_name='traffic_data')
-    op.drop_table('traffic_data')
-    op.drop_index(op.f('ix_traffic_background_jobs_tenant_id'), table_name='traffic_background_jobs')
-    op.drop_index(op.f('ix_traffic_background_jobs_city'), table_name='traffic_background_jobs')
-    op.drop_index('idx_jobs_type_city', table_name='traffic_background_jobs')
-    op.drop_index('idx_jobs_tenant_status', table_name='traffic_background_jobs')
-    op.drop_index('idx_jobs_completed', table_name='traffic_background_jobs')
-    op.drop_index('idx_jobs_city_status', table_name='traffic_background_jobs')
-    op.drop_table('traffic_background_jobs')
-    op.drop_index(op.f('ix_city_weather_data_date'), table_name='city_weather_data')
-    op.drop_index(op.f('ix_city_weather_data_city_id'), table_name='city_weather_data')
-    op.drop_index('idx_city_weather_lookup', table_name='city_weather_data')
-    op.drop_table('city_weather_data')
-    op.drop_index(op.f('ix_city_traffic_data_date'), table_name='city_traffic_data')
-    op.drop_index(op.f('ix_city_traffic_data_city_id'), table_name='city_traffic_data')
-    op.drop_index('idx_city_traffic_lookup', table_name='city_traffic_data')
-    op.drop_table('city_traffic_data')
-    op.drop_index(op.f('ix_audit_logs_user_id'), table_name='audit_logs')
-    op.drop_index(op.f('ix_audit_logs_tenant_id'), table_name='audit_logs')
-    op.drop_index(op.f('ix_audit_logs_severity'), table_name='audit_logs')
-    op.drop_index(op.f('ix_audit_logs_service_name'), table_name='audit_logs')
-    op.drop_index(op.f('ix_audit_logs_resource_type'), table_name='audit_logs')
-    op.drop_index(op.f('ix_audit_logs_resource_id'), table_name='audit_logs')
-    op.drop_index(op.f('ix_audit_logs_created_at'), table_name='audit_logs')
-    op.drop_index(op.f('ix_audit_logs_action'), table_name='audit_logs')
-    op.drop_index('idx_audit_user_created', table_name='audit_logs')
-    op.drop_index('idx_audit_tenant_created', table_name='audit_logs')
-    op.drop_index('idx_audit_severity_created', table_name='audit_logs')
-    op.drop_index('idx_audit_service_created', table_name='audit_logs')
-    op.drop_index('idx_audit_resource_type_action', table_name='audit_logs')
-    op.drop_table('audit_logs')
-    # ### end Alembic commands ###

services/external/migrations/versions/20251102_0856_693e0d98eaf9_add_school_calendars_and_location_.py

@@ -1,69 +0,0 @@
-"""add_school_calendars_and_location_context
-
-Revision ID: 693e0d98eaf9
-Revises: b97bab14ac47
-Create Date: 2025-11-02 08:56:45.463138+01:00
-
-"""
-from typing import Sequence, Union
-
-from alembic import op
-import sqlalchemy as sa
-from sqlalchemy.dialects import postgresql
-
-# revision identifiers, used by Alembic.
-revision: str = '693e0d98eaf9'
-down_revision: Union[str, None] = 'b97bab14ac47'
-branch_labels: Union[str, Sequence[str], None] = None
-depends_on: Union[str, Sequence[str], None] = None
-
-
-def upgrade() -> None:
-    # Create school_calendars table
-    op.create_table(
-        'school_calendars',
-        sa.Column('id', postgresql.UUID(as_uuid=True), nullable=False),
-        sa.Column('city_id', sa.String(length=50), nullable=False),
-        sa.Column('calendar_name', sa.String(length=100), nullable=False),
-        sa.Column('school_type', sa.String(length=20), nullable=False),
-        sa.Column('academic_year', sa.String(length=10), nullable=False),
-        sa.Column('holiday_periods', postgresql.JSONB(astext_type=sa.Text()), nullable=False),
-        sa.Column('school_hours', postgresql.JSONB(astext_type=sa.Text()), nullable=False),
-        sa.Column('source', sa.String(length=100), nullable=True),
-        sa.Column('enabled', sa.Boolean(), nullable=False),
-        sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
-        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
-        sa.PrimaryKeyConstraint('id')
-    )
-    op.create_index('idx_school_calendar_city_year', 'school_calendars', ['city_id', 'academic_year'], unique=False)
-    op.create_index('idx_school_calendar_city_type', 'school_calendars', ['city_id', 'school_type'], unique=False)
-    op.create_index(op.f('ix_school_calendars_city_id'), 'school_calendars', ['city_id'], unique=False)
-
-    # Create tenant_location_contexts table
-    op.create_table(
-        'tenant_location_contexts',
-        sa.Column('tenant_id', postgresql.UUID(as_uuid=True), nullable=False),
-        sa.Column('city_id', sa.String(length=50), nullable=False),
-        sa.Column('school_calendar_id', postgresql.UUID(as_uuid=True), nullable=True),
-        sa.Column('neighborhood', sa.String(length=100), nullable=True),
-        sa.Column('local_events', postgresql.JSONB(astext_type=sa.Text()), nullable=True),
-        sa.Column('notes', sa.String(length=500), nullable=True),
-        sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
-        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
-        sa.PrimaryKeyConstraint('tenant_id')
-    )
-    op.create_index('idx_tenant_location_calendar', 'tenant_location_contexts', ['school_calendar_id'], unique=False)
-    op.create_index(op.f('ix_tenant_location_contexts_city_id'), 'tenant_location_contexts', ['city_id'], unique=False)
-
-
-def downgrade() -> None:
-    # Drop tenant_location_contexts table
-    op.drop_index(op.f('ix_tenant_location_contexts_city_id'), table_name='tenant_location_contexts')
-    op.drop_index('idx_tenant_location_calendar', table_name='tenant_location_contexts')
-    op.drop_table('tenant_location_contexts')
-
-    # Drop school_calendars table
-    op.drop_index(op.f('ix_school_calendars_city_id'), table_name='school_calendars')
-    op.drop_index('idx_school_calendar_city_type', table_name='school_calendars')
-    op.drop_index('idx_school_calendar_city_year', table_name='school_calendars')
-    op.drop_table('school_calendars')

services/external/migrations/versions/20251110_1900_unified_initial_schema.py

@@ -0,0 +1,464 @@
+"""unified_initial_schema
+
+Revision ID: 00001
+Revises:
+Create Date: 2025-11-10 19:00:00.000000+01:00
+
+Complete unified initial schema for External Service including:
+- Weather data collection (weather_data, weather_forecasts, city_weather_data)
+- Traffic data collection (traffic_data, traffic_measurement_points, traffic_background_jobs, city_traffic_data)
+- School calendars and location context (school_calendars, tenant_location_contexts)
+- POI detection system (tenant_poi_contexts, poi_refresh_jobs)
+- Audit logging (audit_logs)
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects.postgresql import UUID, JSONB
+
+
+# revision identifiers, used by Alembic.
+revision: str = '00001'
+down_revision: Union[str, None] = None
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    """Create all tables for External Service"""
+
+    # ============================================================================
+    # AUDIT LOGS
+    # ============================================================================
+    op.create_table(
+        'audit_logs',
+        sa.Column('id', UUID(as_uuid=True), nullable=False),
+        sa.Column('tenant_id', UUID(as_uuid=True), nullable=False),
+        sa.Column('user_id', UUID(as_uuid=True), nullable=False),
+        sa.Column('action', sa.String(length=100), nullable=False),
+        sa.Column('resource_type', sa.String(length=100), nullable=False),
+        sa.Column('resource_id', sa.String(length=255), nullable=True),
+        sa.Column('severity', sa.String(length=20), nullable=False),
+        sa.Column('service_name', sa.String(length=100), nullable=False),
+        sa.Column('description', sa.Text(), nullable=True),
+        sa.Column('changes', JSONB, nullable=True),
+        sa.Column('audit_metadata', JSONB, nullable=True),
+        sa.Column('ip_address', sa.String(length=45), nullable=True),
+        sa.Column('user_agent', sa.Text(), nullable=True),
+        sa.Column('endpoint', sa.String(length=255), nullable=True),
+        sa.Column('method', sa.String(length=10), nullable=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), nullable=False),
+        sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index('idx_audit_resource_type_action', 'audit_logs', ['resource_type', 'action'])
+    op.create_index('idx_audit_service_created', 'audit_logs', ['service_name', 'created_at'])
+    op.create_index('idx_audit_severity_created', 'audit_logs', ['severity', 'created_at'])
+    op.create_index('idx_audit_tenant_created', 'audit_logs', ['tenant_id', 'created_at'])
+    op.create_index('idx_audit_user_created', 'audit_logs', ['user_id', 'created_at'])
+    op.create_index(op.f('ix_audit_logs_action'), 'audit_logs', ['action'])
+    op.create_index(op.f('ix_audit_logs_created_at'), 'audit_logs', ['created_at'])
+    op.create_index(op.f('ix_audit_logs_resource_id'), 'audit_logs', ['resource_id'])
+    op.create_index(op.f('ix_audit_logs_resource_type'), 'audit_logs', ['resource_type'])
+    op.create_index(op.f('ix_audit_logs_service_name'), 'audit_logs', ['service_name'])
+    op.create_index(op.f('ix_audit_logs_severity'), 'audit_logs', ['severity'])
+    op.create_index(op.f('ix_audit_logs_tenant_id'), 'audit_logs', ['tenant_id'])
+    op.create_index(op.f('ix_audit_logs_user_id'), 'audit_logs', ['user_id'])
+
+    # ============================================================================
+    # WEATHER DATA
+    # ============================================================================
+    op.create_table(
+        'city_weather_data',
+        sa.Column('id', UUID(as_uuid=True), nullable=False),
+        sa.Column('city_id', sa.String(length=50), nullable=False),
+        sa.Column('date', sa.DateTime(timezone=True), nullable=False),
+        sa.Column('temperature', sa.Float(), nullable=True),
+        sa.Column('precipitation', sa.Float(), nullable=True),
+        sa.Column('humidity', sa.Float(), nullable=True),
+        sa.Column('wind_speed', sa.Float(), nullable=True),
+        sa.Column('pressure', sa.Float(), nullable=True),
+        sa.Column('description', sa.String(length=200), nullable=True),
+        sa.Column('source', sa.String(length=50), nullable=False),
+        sa.Column('raw_data', JSONB, nullable=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
+        sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index('idx_city_weather_lookup', 'city_weather_data', ['city_id', 'date'])
+    op.create_index(op.f('ix_city_weather_data_city_id'), 'city_weather_data', ['city_id'])
+    op.create_index(op.f('ix_city_weather_data_date'), 'city_weather_data', ['date'])
+
+    op.create_table(
+        'weather_data',
+        sa.Column('id', UUID(as_uuid=True), nullable=False),
+        sa.Column('location_id', sa.String(length=100), nullable=False),
+        sa.Column('city', sa.String(length=50), nullable=False),
+        sa.Column('station_name', sa.String(length=200), nullable=True),
+        sa.Column('latitude', sa.Float(), nullable=True),
+        sa.Column('longitude', sa.Float(), nullable=True),
+        sa.Column('date', sa.DateTime(timezone=True), nullable=False),
+        sa.Column('forecast_date', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('temperature', sa.Float(), nullable=True),
+        sa.Column('temperature_min', sa.Float(), nullable=True),
+        sa.Column('temperature_max', sa.Float(), nullable=True),
+        sa.Column('feels_like', sa.Float(), nullable=True),
+        sa.Column('precipitation', sa.Float(), nullable=True),
+        sa.Column('precipitation_probability', sa.Float(), nullable=True),
+        sa.Column('humidity', sa.Float(), nullable=True),
+        sa.Column('wind_speed', sa.Float(), nullable=True),
+        sa.Column('wind_direction', sa.Float(), nullable=True),
+        sa.Column('wind_gust', sa.Float(), nullable=True),
+        sa.Column('pressure', sa.Float(), nullable=True),
+        sa.Column('visibility', sa.Float(), nullable=True),
+        sa.Column('uv_index', sa.Float(), nullable=True),
+        sa.Column('cloud_cover', sa.Float(), nullable=True),
+        sa.Column('condition', sa.String(length=100), nullable=True),
+        sa.Column('description', sa.String(length=200), nullable=True),
+        sa.Column('weather_code', sa.String(length=20), nullable=True),
+        sa.Column('source', sa.String(length=50), nullable=False),
+        sa.Column('data_type', sa.String(length=20), nullable=False),
+        sa.Column('is_forecast', sa.Boolean(), nullable=True),
+        sa.Column('data_quality_score', sa.Float(), nullable=True),
+        sa.Column('raw_data', JSONB, nullable=True),
+        sa.Column('processed_data', JSONB, nullable=True),
+        sa.Column('tenant_id', UUID(as_uuid=True), nullable=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
+        sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index('idx_weather_location_date', 'weather_data', ['location_id', 'date'])
+    op.create_index(op.f('ix_weather_data_date'), 'weather_data', ['date'])
+    op.create_index(op.f('ix_weather_data_location_id'), 'weather_data', ['location_id'])
+    op.create_index(op.f('ix_weather_data_tenant_id'), 'weather_data', ['tenant_id'])
+
+    op.create_table(
+        'weather_forecasts',
+        sa.Column('id', UUID(as_uuid=True), nullable=False),
+        sa.Column('location_id', sa.String(length=100), nullable=False),
+        sa.Column('forecast_date', sa.DateTime(timezone=True), nullable=False),
+        sa.Column('generated_at', sa.DateTime(timezone=True), nullable=False),
+        sa.Column('temperature', sa.Float(), nullable=True),
+        sa.Column('precipitation', sa.Float(), nullable=True),
+        sa.Column('humidity', sa.Float(), nullable=True),
+        sa.Column('wind_speed', sa.Float(), nullable=True),
+        sa.Column('description', sa.String(length=200), nullable=True),
+        sa.Column('source', sa.String(length=50), nullable=False),
+        sa.Column('raw_data', sa.Text(), nullable=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
+        sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index('idx_forecast_location_date', 'weather_forecasts', ['location_id', 'forecast_date'])
+    op.create_index(op.f('ix_weather_forecasts_location_id'), 'weather_forecasts', ['location_id'])
+
+    # ============================================================================
+    # TRAFFIC DATA
+    # ============================================================================
+    op.create_table(
+        'city_traffic_data',
+        sa.Column('id', UUID(as_uuid=True), nullable=False),
+        sa.Column('city_id', sa.String(length=50), nullable=False),
+        sa.Column('date', sa.DateTime(timezone=True), nullable=False),
+        sa.Column('traffic_volume', sa.Integer(), nullable=True),
+        sa.Column('pedestrian_count', sa.Integer(), nullable=True),
+        sa.Column('congestion_level', sa.String(length=20), nullable=True),
+        sa.Column('average_speed', sa.Float(), nullable=True),
+        sa.Column('source', sa.String(length=50), nullable=False),
+        sa.Column('raw_data', JSONB, nullable=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
+        sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index('idx_city_traffic_lookup', 'city_traffic_data', ['city_id', 'date'])
+    op.create_index(op.f('ix_city_traffic_data_city_id'), 'city_traffic_data', ['city_id'])
+    op.create_index(op.f('ix_city_traffic_data_date'), 'city_traffic_data', ['date'])
+
+    op.create_table(
+        'traffic_measurement_points',
+        sa.Column('id', UUID(as_uuid=True), nullable=False),
+        sa.Column('city', sa.String(length=50), nullable=False),
+        sa.Column('measurement_point_id', sa.String(length=100), nullable=False),
+        sa.Column('name', sa.String(length=500), nullable=True),
+        sa.Column('description', sa.Text(), nullable=True),
+        sa.Column('latitude', sa.Float(), nullable=False),
+        sa.Column('longitude', sa.Float(), nullable=False),
+        sa.Column('district', sa.String(length=100), nullable=True),
+        sa.Column('zone', sa.String(length=100), nullable=True),
+        sa.Column('road_type', sa.String(length=50), nullable=True),
+        sa.Column('measurement_type', sa.String(length=50), nullable=True),
+        sa.Column('point_category', sa.String(length=50), nullable=True),
+        sa.Column('is_active', sa.Boolean(), nullable=True),
+        sa.Column('installation_date', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('last_data_received', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('data_quality_rating', sa.Float(), nullable=True),
+        sa.Column('city_specific_metadata', JSONB, nullable=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
+        sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index('idx_points_active', 'traffic_measurement_points', ['city', 'is_active', 'last_data_received'])
+    op.create_index('idx_points_city_location', 'traffic_measurement_points', ['city', 'latitude', 'longitude'])
+    op.create_index('idx_points_district', 'traffic_measurement_points', ['city', 'district'])
+    op.create_index('idx_points_road_type', 'traffic_measurement_points', ['city', 'road_type'])
+    op.create_index('idx_unique_city_point', 'traffic_measurement_points', ['city', 'measurement_point_id'], unique=True)
+    op.create_index(op.f('ix_traffic_measurement_points_city'), 'traffic_measurement_points', ['city'])
+    op.create_index(op.f('ix_traffic_measurement_points_measurement_point_id'), 'traffic_measurement_points', ['measurement_point_id'])
+
+    op.create_table(
+        'traffic_data',
+        sa.Column('id', UUID(as_uuid=True), nullable=False),
+        sa.Column('location_id', sa.String(length=100), nullable=False),
+        sa.Column('city', sa.String(length=50), nullable=False),
+        sa.Column('date', sa.DateTime(timezone=True), nullable=False),
+        sa.Column('traffic_volume', sa.Integer(), nullable=True),
+        sa.Column('congestion_level', sa.String(length=20), nullable=True),
+        sa.Column('average_speed', sa.Float(), nullable=True),
+        sa.Column('occupation_percentage', sa.Float(), nullable=True),
+        sa.Column('load_percentage', sa.Float(), nullable=True),
+        sa.Column('pedestrian_count', sa.Integer(), nullable=True),
+        sa.Column('measurement_point_id', sa.String(length=100), nullable=True),
+        sa.Column('measurement_point_name', sa.String(length=500), nullable=True),
+        sa.Column('measurement_point_type', sa.String(length=50), nullable=True),
+        sa.Column('latitude', sa.Float(), nullable=True),
+        sa.Column('longitude', sa.Float(), nullable=True),
+        sa.Column('district', sa.String(length=100), nullable=True),
+        sa.Column('zone', sa.String(length=100), nullable=True),
+        sa.Column('source', sa.String(length=50), nullable=False),
+        sa.Column('data_quality_score', sa.Float(), nullable=True),
+        sa.Column('is_synthetic', sa.Boolean(), nullable=True),
+        sa.Column('has_pedestrian_inference', sa.Boolean(), nullable=True),
+        sa.Column('city_specific_data', JSONB, nullable=True),
+        sa.Column('raw_data', sa.Text(), nullable=True),
+        sa.Column('tenant_id', UUID(as_uuid=True), nullable=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
+        sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index('idx_traffic_city_date', 'traffic_data', ['city', 'date'])
+    op.create_index('idx_traffic_city_location', 'traffic_data', ['city', 'location_id'])
+    op.create_index('idx_traffic_district_date', 'traffic_data', ['city', 'district', 'date'])
+    op.create_index('idx_traffic_location_date', 'traffic_data', ['location_id', 'date'])
+    op.create_index('idx_traffic_measurement_point', 'traffic_data', ['city', 'measurement_point_id'])
+    op.create_index('idx_traffic_quality', 'traffic_data', ['city', 'data_quality_score', 'date'])
+    op.create_index('idx_traffic_tenant_date', 'traffic_data', ['tenant_id', 'date'])
+    op.create_index('idx_traffic_training', 'traffic_data', ['tenant_id', 'city', 'date', 'is_synthetic'])
+    op.create_index(op.f('ix_traffic_data_city'), 'traffic_data', ['city'])
+    op.create_index(op.f('ix_traffic_data_date'), 'traffic_data', ['date'])
+    op.create_index(op.f('ix_traffic_data_location_id'), 'traffic_data', ['location_id'])
+    op.create_index(op.f('ix_traffic_data_measurement_point_id'), 'traffic_data', ['measurement_point_id'])
+    op.create_index(op.f('ix_traffic_data_tenant_id'), 'traffic_data', ['tenant_id'])
+
+    op.create_table(
+        'traffic_background_jobs',
+        sa.Column('id', UUID(as_uuid=True), nullable=False),
+        sa.Column('job_type', sa.String(length=50), nullable=False),
+        sa.Column('city', sa.String(length=50), nullable=False),
+        sa.Column('location_pattern', sa.String(length=200), nullable=True),
+        sa.Column('scheduled_at', sa.DateTime(timezone=True), nullable=False),
+        sa.Column('started_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('completed_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('status', sa.String(length=20), nullable=False),
+        sa.Column('progress_percentage', sa.Float(), nullable=True),
+        sa.Column('records_processed', sa.Integer(), nullable=True),
+        sa.Column('records_stored', sa.Integer(), nullable=True),
+        sa.Column('data_start_date', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('data_end_date', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('success_count', sa.Integer(), nullable=True),
+        sa.Column('error_count', sa.Integer(), nullable=True),
+        sa.Column('error_message', sa.Text(), nullable=True),
+        sa.Column('job_metadata', JSONB, nullable=True),
+        sa.Column('tenant_id', UUID(as_uuid=True), nullable=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
+        sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index('idx_jobs_city_status', 'traffic_background_jobs', ['city', 'status', 'scheduled_at'])
+    op.create_index('idx_jobs_completed', 'traffic_background_jobs', ['status', 'completed_at'])
+    op.create_index('idx_jobs_tenant_status', 'traffic_background_jobs', ['tenant_id', 'status', 'scheduled_at'])
+    op.create_index('idx_jobs_type_city', 'traffic_background_jobs', ['job_type', 'city', 'scheduled_at'])
+    op.create_index(op.f('ix_traffic_background_jobs_city'), 'traffic_background_jobs', ['city'])
+    op.create_index(op.f('ix_traffic_background_jobs_tenant_id'), 'traffic_background_jobs', ['tenant_id'])
+
+    # ============================================================================
+    # SCHOOL CALENDARS & LOCATION CONTEXT
+    # ============================================================================
+    op.create_table(
+        'school_calendars',
+        sa.Column('id', UUID(as_uuid=True), nullable=False),
+        sa.Column('city_id', sa.String(length=50), nullable=False),
+        sa.Column('calendar_name', sa.String(length=100), nullable=False),
+        sa.Column('school_type', sa.String(length=20), nullable=False),
+        sa.Column('academic_year', sa.String(length=10), nullable=False),
+        sa.Column('holiday_periods', JSONB, nullable=False),
+        sa.Column('school_hours', JSONB, nullable=False),
+        sa.Column('source', sa.String(length=100), nullable=True),
+        sa.Column('enabled', sa.Boolean(), nullable=False),
+        sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
+        sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index('idx_school_calendar_city_year', 'school_calendars', ['city_id', 'academic_year'])
+    op.create_index('idx_school_calendar_city_type', 'school_calendars', ['city_id', 'school_type'])
+    op.create_index(op.f('ix_school_calendars_city_id'), 'school_calendars', ['city_id'])
+
+    op.create_table(
+        'tenant_location_contexts',
+        sa.Column('tenant_id', UUID(as_uuid=True), nullable=False),
+        sa.Column('city_id', sa.String(length=50), nullable=False),
+        sa.Column('school_calendar_id', UUID(as_uuid=True), nullable=True),
+        sa.Column('neighborhood', sa.String(length=100), nullable=True),
+        sa.Column('local_events', JSONB, nullable=True),
+        sa.Column('notes', sa.String(length=500), nullable=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=True),
+        sa.PrimaryKeyConstraint('tenant_id')
+    )
+    op.create_index('idx_tenant_location_calendar', 'tenant_location_contexts', ['school_calendar_id'])
+    op.create_index(op.f('ix_tenant_location_contexts_city_id'), 'tenant_location_contexts', ['city_id'])
+
+    # ============================================================================
+    # POI DETECTION SYSTEM
+    # ============================================================================
+    op.create_table(
+        'tenant_poi_contexts',
+        sa.Column('id', UUID(as_uuid=True), primary_key=True),
+        sa.Column('tenant_id', UUID(as_uuid=True), nullable=False, unique=True, index=True),
+        sa.Column('latitude', sa.Float(), nullable=False),
+        sa.Column('longitude', sa.Float(), nullable=False),
+        sa.Column('poi_detection_results', JSONB, nullable=False, server_default='{}'),
+        sa.Column('ml_features', JSONB, nullable=False, server_default='{}'),
+        sa.Column('total_pois_detected', sa.Integer(), default=0),
+        sa.Column('high_impact_categories', JSONB, server_default='[]'),
+        sa.Column('relevant_categories', JSONB, server_default='[]'),
+        sa.Column('detection_timestamp', sa.DateTime(timezone=True), nullable=False),
+        sa.Column('detection_source', sa.String(50), default='overpass_api'),
+        sa.Column('detection_status', sa.String(20), default='completed'),
+        sa.Column('detection_error', sa.String(500), nullable=True),
+        sa.Column('next_refresh_date', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('refresh_interval_days', sa.Integer(), default=180),
+        sa.Column('last_refreshed_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), server_default=sa.func.now()),
+        sa.Column('updated_at', sa.DateTime(timezone=True), server_default=sa.func.now(), onupdate=sa.func.now())
+    )
+    op.create_index('idx_tenant_poi_location', 'tenant_poi_contexts', ['latitude', 'longitude'])
+    op.create_index('idx_tenant_poi_refresh', 'tenant_poi_contexts', ['next_refresh_date'])
+    op.create_index('idx_tenant_poi_status', 'tenant_poi_contexts', ['detection_status'])
+
+    op.create_table(
+        'poi_refresh_jobs',
+        sa.Column('id', UUID(as_uuid=True), primary_key=True),
+        sa.Column('tenant_id', UUID(as_uuid=True), nullable=False, index=True),
+        sa.Column('scheduled_at', sa.DateTime(timezone=True), nullable=False, index=True),
+        sa.Column('started_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('completed_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('status', sa.String(50), nullable=False, default='pending', index=True),
+        sa.Column('attempt_count', sa.Integer, nullable=False, default=0),
+        sa.Column('max_attempts', sa.Integer, nullable=False, default=3),
+        sa.Column('latitude', sa.Float, nullable=False),
+        sa.Column('longitude', sa.Float, nullable=False),
+        sa.Column('pois_detected', sa.Integer, nullable=True),
+        sa.Column('changes_detected', sa.Boolean, default=False),
+        sa.Column('change_summary', JSONB, nullable=True),
+        sa.Column('error_message', sa.Text, nullable=True),
+        sa.Column('error_details', JSONB, nullable=True),
+        sa.Column('next_scheduled_at', sa.DateTime(timezone=True), nullable=True, index=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), nullable=False, server_default=sa.func.now()),
+        sa.Column('updated_at', sa.DateTime(timezone=True), nullable=False, server_default=sa.func.now(), onupdate=sa.func.now())
+    )
+    op.create_index('idx_poi_refresh_jobs_tenant_status', 'poi_refresh_jobs', ['tenant_id', 'status'])
+    op.create_index('idx_poi_refresh_jobs_status_scheduled', 'poi_refresh_jobs', ['status', 'scheduled_at'])
+
+
+def downgrade() -> None:
+    """Drop all tables"""
+
+    # POI Detection System
+    op.drop_index('idx_poi_refresh_jobs_status_scheduled', table_name='poi_refresh_jobs')
+    op.drop_index('idx_poi_refresh_jobs_tenant_status', table_name='poi_refresh_jobs')
+    op.drop_table('poi_refresh_jobs')
+
+    op.drop_index('idx_tenant_poi_status', table_name='tenant_poi_contexts')
+    op.drop_index('idx_tenant_poi_refresh', table_name='tenant_poi_contexts')
+    op.drop_index('idx_tenant_poi_location', table_name='tenant_poi_contexts')
+    op.drop_table('tenant_poi_contexts')
+
+    # School Calendars & Location Context
+    op.drop_index(op.f('ix_tenant_location_contexts_city_id'), table_name='tenant_location_contexts')
+    op.drop_index('idx_tenant_location_calendar', table_name='tenant_location_contexts')
+    op.drop_table('tenant_location_contexts')
+
+    op.drop_index(op.f('ix_school_calendars_city_id'), table_name='school_calendars')
+    op.drop_index('idx_school_calendar_city_type', table_name='school_calendars')
+    op.drop_index('idx_school_calendar_city_year', table_name='school_calendars')
+    op.drop_table('school_calendars')
+
+    # Traffic Data
+    op.drop_index(op.f('ix_traffic_background_jobs_tenant_id'), table_name='traffic_background_jobs')
+    op.drop_index(op.f('ix_traffic_background_jobs_city'), table_name='traffic_background_jobs')
+    op.drop_index('idx_jobs_type_city', table_name='traffic_background_jobs')
+    op.drop_index('idx_jobs_tenant_status', table_name='traffic_background_jobs')
+    op.drop_index('idx_jobs_completed', table_name='traffic_background_jobs')
+    op.drop_index('idx_jobs_city_status', table_name='traffic_background_jobs')
+    op.drop_table('traffic_background_jobs')
+
+    op.drop_index(op.f('ix_traffic_data_tenant_id'), table_name='traffic_data')
+    op.drop_index(op.f('ix_traffic_data_measurement_point_id'), table_name='traffic_data')
+    op.drop_index(op.f('ix_traffic_data_location_id'), table_name='traffic_data')
+    op.drop_index(op.f('ix_traffic_data_date'), table_name='traffic_data')
+    op.drop_index(op.f('ix_traffic_data_city'), table_name='traffic_data')
+    op.drop_index('idx_traffic_training', table_name='traffic_data')
+    op.drop_index('idx_traffic_tenant_date', table_name='traffic_data')
+    op.drop_index('idx_traffic_quality', table_name='traffic_data')
+    op.drop_index('idx_traffic_measurement_point', table_name='traffic_data')
+    op.drop_index('idx_traffic_location_date', table_name='traffic_data')
+    op.drop_index('idx_traffic_district_date', table_name='traffic_data')
+    op.drop_index('idx_traffic_city_location', table_name='traffic_data')
+    op.drop_index('idx_traffic_city_date', table_name='traffic_data')
+    op.drop_table('traffic_data')
+
+    op.drop_index(op.f('ix_traffic_measurement_points_measurement_point_id'), table_name='traffic_measurement_points')
+    op.drop_index(op.f('ix_traffic_measurement_points_city'), table_name='traffic_measurement_points')
+    op.drop_index('idx_unique_city_point', table_name='traffic_measurement_points')
+    op.drop_index('idx_points_road_type', table_name='traffic_measurement_points')
+    op.drop_index('idx_points_district', table_name='traffic_measurement_points')
+    op.drop_index('idx_points_city_location', table_name='traffic_measurement_points')
+    op.drop_index('idx_points_active', table_name='traffic_measurement_points')
+    op.drop_table('traffic_measurement_points')
+
+    op.drop_index(op.f('ix_city_traffic_data_date'), table_name='city_traffic_data')
+    op.drop_index(op.f('ix_city_traffic_data_city_id'), table_name='city_traffic_data')
+    op.drop_index('idx_city_traffic_lookup', table_name='city_traffic_data')
+    op.drop_table('city_traffic_data')
+
+    # Weather Data
+    op.drop_index(op.f('ix_weather_forecasts_location_id'), table_name='weather_forecasts')
+    op.drop_index('idx_forecast_location_date', table_name='weather_forecasts')
+    op.drop_table('weather_forecasts')
+
+    op.drop_index(op.f('ix_weather_data_tenant_id'), table_name='weather_data')
+    op.drop_index(op.f('ix_weather_data_location_id'), table_name='weather_data')
+    op.drop_index(op.f('ix_weather_data_date'), table_name='weather_data')
+    op.drop_index('idx_weather_location_date', table_name='weather_data')
+    op.drop_table('weather_data')
+
+    op.drop_index(op.f('ix_city_weather_data_date'), table_name='city_weather_data')
+    op.drop_index(op.f('ix_city_weather_data_city_id'), table_name='city_weather_data')
+    op.drop_index('idx_city_weather_lookup', table_name='city_weather_data')
+    op.drop_table('city_weather_data')
+
+    # Audit Logs
+    op.drop_index(op.f('ix_audit_logs_user_id'), table_name='audit_logs')
+    op.drop_index(op.f('ix_audit_logs_tenant_id'), table_name='audit_logs')
+    op.drop_index(op.f('ix_audit_logs_severity'), table_name='audit_logs')
+    op.drop_index(op.f('ix_audit_logs_service_name'), table_name='audit_logs')
+    op.drop_index(op.f('ix_audit_logs_resource_type'), table_name='audit_logs')
+    op.drop_index(op.f('ix_audit_logs_resource_id'), table_name='audit_logs')
+    op.drop_index(op.f('ix_audit_logs_created_at'), table_name='audit_logs')
+    op.drop_index(op.f('ix_audit_logs_action'), table_name='audit_logs')
+    op.drop_index('idx_audit_user_created', table_name='audit_logs')
+    op.drop_index('idx_audit_tenant_created', table_name='audit_logs')
+    op.drop_index('idx_audit_severity_created', table_name='audit_logs')
+    op.drop_index('idx_audit_service_created', table_name='audit_logs')
+    op.drop_index('idx_audit_resource_type_action', table_name='audit_logs')
+    op.drop_table('audit_logs')

services/external/requirements.txt

@@ -48,6 +48,9 @@ lxml==5.3.0
 # Geospatial processing
 pyproj==3.7.1
 
+# OpenStreetMap / POI detection
+overpy==0.7
+
 # Development
 python-multipart==0.0.6