Merge branch 'main' of github.com:SirBlobby/VTHacks13

llm/flask_server.py

@@ -366,6 +366,87 @@ def get_single_route_endpoint():
             'error': str(e)
         }), 500
 
+@app.route('/predict', methods=['POST', 'OPTIONS'])
+def predict_crash_magnitude():
+    """
+    Predict crash magnitude for a route using AI model.
+    Expected request body:
+    {
+        "source": {"lat": float, "lon": float},
+        "destination": {"lat": float, "lon": float}
+    }
+    """
+    # Handle preflight CORS request
+    if request.method == 'OPTIONS':
+        response = jsonify({'status': 'ok'})
+        response.headers.add('Access-Control-Allow-Origin', '*')
+        response.headers.add('Access-Control-Allow-Headers', 'Content-Type')
+        response.headers.add('Access-Control-Allow-Methods', 'POST, OPTIONS')
+        return response
+    
+    try:
+        data = request.get_json()
+        if not data:
+            return jsonify({'error': 'No JSON data provided'}), 400
+            
+        # Validate required fields
+        if 'source' not in data or 'destination' not in data:
+            return jsonify({'error': 'Missing source or destination coordinates'}), 400
+            
+        source = data['source']
+        destination = data['destination']
+        
+        # Validate coordinate format
+        required_fields = ['lat', 'lon']
+        for coord_set, name in [(source, 'source'), (destination, 'destination')]:
+            for field in required_fields:
+                if field not in coord_set:
+                    return jsonify({'error': f'Missing {field} in {name} coordinates'}), 400
+                try:
+                    float(coord_set[field])
+                except (TypeError, ValueError):
+                    return jsonify({'error': f'Invalid {field} value in {name} coordinates'}), 400
+        
+        # For now, return a mock prediction based on distance
+        # In a real implementation, this would call your AI model
+        import math
+        
+        lat1, lon1 = float(source['lat']), float(source['lon'])
+        lat2, lon2 = float(destination['lat']), float(destination['lon'])
+        
+        # Calculate distance (rough approximation)
+        distance = math.sqrt((lat2 - lat1)**2 + (lon2 - lon1)**2)
+        
+        # Mock prediction: longer routes might have higher crash magnitude
+        # This is just placeholder logic until you integrate your actual AI model
+        base_magnitude = min(distance * 50, 1.0)  # Cap at 1.0
+        confidence = 0.85  # Mock confidence
+        
+        response_data = {
+            'prediction': {
+                'prediction': base_magnitude,
+                'confidence': confidence
+            },
+            'called_with': f"Route from ({lat1}, {lon1}) to ({lat2}, {lon2})",
+            'diagnostics': {
+                'input_dim': 4  # lat1, lon1, lat2, lon2
+            }
+        }
+        
+        print(f"🔮 Crash magnitude prediction request: {data}")
+        print(f"📊 Returning prediction: {response_data['prediction']['prediction']:.3f}")
+        
+        response = jsonify(response_data)
+        response.headers.add('Access-Control-Allow-Origin', '*')
+        return response
+        
+    except Exception as e:
+        print(f"❌ Error in crash magnitude prediction: {e}")
+        traceback.print_exc()
+        error_response = jsonify({'error': str(e)})
+        error_response.headers.add('Access-Control-Allow-Origin', '*')
+        return error_response, 500
+
 @app.errorhandler(404)
 def not_found(error):
     return jsonify({'success': False, 'error': 'Endpoint not found'}), 404
@@ -382,6 +463,7 @@ if __name__ == '__main__':
     print("   - POST /api/analyze-crashes")
     print("   - POST /api/find-safe-route")
     print("   - POST /api/get-single-route")
+    print("   - POST /predict (AI crash magnitude prediction)")
     print("\n🌐 Server running on http://localhost:5001")
     
     app.run(debug=True, host='0.0.0.0', port=5001)

roadcast/app.py

@@ -1,4 +1,5 @@
 from flask import Flask, request, jsonify
+from flask_cors import CORS
 from dotenv import load_dotenv
 from train import compute_index
 from models import load_model
@@ -8,6 +9,14 @@ from models import MLP
 load_dotenv()
 
 app = Flask(__name__)
+# Enable CORS for all routes, origins, and methods
+CORS(app, resources={
+    r"/*": {
+        "origins": "*",
+        "methods": ["GET", "POST", "PUT", "DELETE", "OPTIONS"],
+        "allow_headers": ["Content-Type", "Authorization", "Accept", "Origin", "X-Requested-With"]
+    }
+})
 import os
 import threading
 import json
@@ -121,6 +130,7 @@ def predict_endpoint():
     # build feature vector of correct length and populate lat/lon using preprocess meta if available
     feature_vector = np.zeros(int(input_dim), dtype=float)
     meta_path = os.path.join(os.getcwd(), 'preprocess_meta.npz')
+    
     if os.path.exists(meta_path):
         try:
             meta = np.load(meta_path, allow_pickle=True)
@@ -128,33 +138,83 @@ def predict_endpoint():
             means = meta.get('means')
             if means is not None and len(means) == input_dim:
                 feature_vector[:] = means
+                
             col_lower = [c.lower() for c in cols]
-            if 'lat' in col_lower:
+            print(f"📋 Available columns: {col_lower[:10]}...")  # Show first 10 columns
-                feature_vector[col_lower.index('lat')] = src_lat
+            
-            elif 'latitude' in col_lower:
+            # Try to find and populate coordinate fields
-                feature_vector[col_lower.index('latitude')] = src_lat
+            coord_mappings = [
-            else:
+                (('lat', 'latitude', 'src_lat', 'source_lat'), src_lat),
-                feature_vector[0] = src_lat
+                (('lon', 'lng', 'longitude', 'src_lon', 'source_lon'), src_lon),
-            if 'lon' in col_lower:
+                (('dst_lat', 'dest_lat', 'destination_lat', 'end_lat'), dst_lat),
-                feature_vector[col_lower.index('lon')] = src_lon
+                (('dst_lon', 'dest_lon', 'destination_lon', 'end_lon', 'dst_lng'), dst_lon)
-            elif 'longitude' in col_lower:
+            ]
-                feature_vector[col_lower.index('longitude')] = src_lon
+            
-            else:
+            for possible_names, value in coord_mappings:
-                if input_dim > 1:
+                for name in possible_names:
-                    feature_vector[1] = src_lon
+                    if name in col_lower:
-        except Exception:
+                        idx = col_lower.index(name)
+                        feature_vector[idx] = value
+                        print(f"✅ Mapped {name} (index {idx}) = {value}")
+                        break
+                        
+            # Calculate route features that might be useful
+            route_distance = ((dst_lat - src_lat)**2 + (dst_lon - src_lon)**2)**0.5
+            midpoint_lat = (src_lat + dst_lat) / 2
+            midpoint_lon = (src_lon + dst_lon) / 2
+            
+            # Try to populate additional features that might exist
+            additional_features = {
+                'distance': route_distance,
+                'route_distance': route_distance,
+                'midpoint_lat': midpoint_lat,
+                'midpoint_lon': midpoint_lon,
+                'lat_diff': abs(dst_lat - src_lat),
+                'lon_diff': abs(dst_lon - src_lon)
+            }
+            
+            for feature_name, feature_value in additional_features.items():
+                if feature_name in col_lower:
+                    idx = col_lower.index(feature_name)
+                    feature_vector[idx] = feature_value
+                    print(f"✅ Mapped {feature_name} (index {idx}) = {feature_value}")
+                    
+        except Exception as e:
+            print(f"⚠️ Error processing metadata: {e}")
+            # Fallback to simple coordinate mapping
             feature_vector[:] = 0.0
             feature_vector[0] = src_lat
             if input_dim > 1:
                 feature_vector[1] = src_lon
+            if input_dim > 2:
+                feature_vector[2] = dst_lat
+            if input_dim > 3:
+                feature_vector[3] = dst_lon
     else:
+        print("⚠️ No preprocess_meta.npz found, using simple coordinate mapping")
+        # Simple fallback mapping
         feature_vector[0] = src_lat
         if input_dim > 1:
             feature_vector[1] = src_lon
+        if input_dim > 2:
+            feature_vector[2] = dst_lat
+        if input_dim > 3:
+            feature_vector[3] = dst_lon
+        
+        # Add some derived features to create more variation
+        if input_dim > 4:
+            feature_vector[4] = ((dst_lat - src_lat)**2 + (dst_lon - src_lon)**2)**0.5  # distance
+        if input_dim > 5:
+            feature_vector[5] = (src_lat + dst_lat) / 2  # midpoint lat
+        if input_dim > 6:
+            feature_vector[6] = (src_lon + dst_lon) / 2  # midpoint lon
 
     # compute index using model
     try:
+        print(f"🔍 Feature vector for prediction: {feature_vector[:8]}...")  # Show first 8 values
+        print(f"📍 Coordinates: src({src_lat}, {src_lon}) → dst({dst_lat}, {dst_lon})")
         index = compute_index(model, feature_vector)
+        print(f"📊 Computed index: {index}")
     except Exception as e:
         return jsonify({"error": "compute_index failed", "detail": str(e)}), 500
 

web/src/app/components/MapView.tsx

@@ -5,7 +5,7 @@ import mapboxgl from 'mapbox-gl';
 import 'mapbox-gl/dist/mapbox-gl.css';
 import MapboxGeocoder from '@mapbox/mapbox-gl-geocoder';
 import '@mapbox/mapbox-gl-geocoder/dist/mapbox-gl-geocoder.css';
-import { generateDCPoints, haversine, PointFeature, convertCrashDataToGeoJSON } from '../lib/mapUtils';
+import { generateDCPoints, generateDCPointsWithAI, haversine, PointFeature, convertCrashDataToGeoJSON, convertCrashDataToGeoJSONWithAI } from '../lib/mapUtils';
 import { useCrashData, UseCrashDataResult } from '../hooks/useCrashData';
 import { CrashData } from '../api/crashes/route';
 import { WeatherData, CrashAnalysisData } from '../../lib/flaskApi';
@@ -49,6 +49,7 @@ interface MapViewProps {
 	crashData?: CrashData[]; // external crash data to use
 	crashDataHook?: UseCrashDataResult; // the crash data hook from main page
 	isMapPickingMode?: boolean; // whether map is in picking mode (prevents popups)
+	useAIMagnitudes?: boolean; // whether to use AI-predicted crash magnitudes
 }
 
 export default function MapView({ 
@@ -63,7 +64,8 @@ export default function MapView({
 	useRealCrashData = true,
 	crashData = [],
 	crashDataHook,
-	isMapPickingMode = false
+	isMapPickingMode = false,
+	useAIMagnitudes = true // Default to true to use AI predictions
 }: MapViewProps) {
 	const containerRef = useRef<HTMLDivElement | null>(null);
 	const mapContainerRef = useRef<HTMLDivElement | null>(null);
@@ -71,6 +73,7 @@ export default function MapView({
 	const styleChoiceRef = useRef<'dark' | 'streets'>(mapStyleChoice);
 	const isMapPickingModeRef = useRef<boolean>(isMapPickingMode);
 	const [size, setSize] = useState({ width: 0, height: 0 });
+	const [isLoadingAIPredictions, setIsLoadingAIPredictions] = useState(false);
 	const dcDataRef = useRef<GeoJSON.FeatureCollection | null>(null);
 	const internalCrashDataHook = useCrashData({ autoLoad: false, limit: 10000 }); // Don't auto-load if external data provided
 
@@ -84,9 +87,25 @@ export default function MapView({
 		const currentCrashDataHook = crashDataHook || internalCrashDataHook;
 		const activeData = crashData.length > 0 ? crashData : currentCrashDataHook.data;
 		console.log('MapView useEffect: crashData.length =', crashData.length, 'crashDataHook.data.length =', currentCrashDataHook.data.length);
+		
 		if (useRealCrashData && activeData.length > 0) {
 			console.log('Converting crash data to GeoJSON...');
-			dcDataRef.current = convertCrashDataToGeoJSON(activeData);
+			
+			const processData = async () => {
+				setIsLoadingAIPredictions(useAIMagnitudes);
+				
+				let geoJSONData: GeoJSON.FeatureCollection;
+				if (useAIMagnitudes) {
+					console.log('🤖 Using AI-enhanced crash data conversion...');
+					geoJSONData = await convertCrashDataToGeoJSONWithAI(activeData);
+				} else {
+					console.log('📊 Using standard crash data conversion...');
+					geoJSONData = convertCrashDataToGeoJSON(activeData);
+				}
+				
+				dcDataRef.current = geoJSONData;
+				setIsLoadingAIPredictions(false);
+				
 				// Update the map source if map is ready
 				const map = mapRef.current;
 				if (map && map.isStyleLoaded()) {
@@ -94,70 +113,14 @@ export default function MapView({
 					if (map.getSource('dc-quakes')) {
 						(map.getSource('dc-quakes') as mapboxgl.GeoJSONSource).setData(dcDataRef.current);
 					} else {
-					console.log('Source not found, calling addDataAndLayers');
+						console.log('Source not found, will be added when map loads');
-					// Call the inner function manually - we need to recreate it here
-					if (dcDataRef.current) {
-						console.log('Adding data and layers, data has', dcDataRef.current.features.length, 'features');
-						if (!map.getSource('dc-quakes')) {
-							console.log('Creating new source');
-							map.addSource('dc-quakes', { type: 'geojson', data: dcDataRef.current });
-						}
-						// Add layers if they don't exist
-						if (!map.getLayer('dc-heat')) {
-							map.addLayer({
-								id: 'dc-heat', type: 'heatmap', source: 'dc-quakes',
-								paint: {
-									'heatmap-weight': ['interpolate', ['linear'], ['get', 'mag'], 0, 0, 6, 1],
-									'heatmap-intensity': heatIntensity,
-									'heatmap-color': [
-										'interpolate', 
-										['linear'], 
-										['heatmap-density'], 
-										0, 'rgba(0,0,0,0)', 
-										0.2, 'rgba(255,255,0,0.7)', 
-										0.4, 'rgba(255,165,0,0.8)', 
-										0.6, 'rgba(255,69,0,0.9)', 
-										0.8, 'rgba(255,0,0,0.95)', 
-										1, 'rgba(139,0,0,1)'
-									],
-									'heatmap-radius': heatRadius,
-									'heatmap-opacity': ['interpolate', ['linear'], ['zoom'], 7, 1, 12, 0.8]
-								}
-							});
-						}
-						if (!map.getLayer('dc-point')) {
-							map.addLayer({
-								id: 'dc-point', type: 'circle', source: 'dc-quakes', minzoom: 12,
-								paint: {
-									'circle-radius': ['interpolate', ['linear'], ['get', 'mag'], 1, 3, 6, 10],
-									'circle-color': [
-										'interpolate', 
-										['linear'], 
-										['get', 'mag'], 
-										1, styleChoiceRef.current === 'dark' ? '#ffff99' : '#ffa500',
-										3, styleChoiceRef.current === 'dark' ? '#ff6666' : '#ff4500',
-										6, styleChoiceRef.current === 'dark' ? '#ff0000' : '#8b0000'
-									] as any,
-									'circle-opacity': ['interpolate', ['linear'], ['zoom'], 12, 0.7, 14, 0.9],
-									'circle-stroke-width': 1,
-									'circle-stroke-color': styleChoiceRef.current === 'dark' ? '#ffffff' : '#000000'
-								}
-							});
-						}
-						// Update layer visibility
-						if (map.getLayer('dc-heat')) {
-							map.setLayoutProperty('dc-heat', 'visibility', heatVisible ? 'visible' : 'none');
-						}
-						if (map.getLayer('dc-point')) {
-							map.setLayoutProperty('dc-point', 'visibility', pointsVisible ? 'visible' : 'none');
 					}
 				}
+			};
+			
+			processData().catch(console.error);
 		}
-			} else {
+	}, [crashData, crashDataHook, useRealCrashData, useAIMagnitudes]);
-				console.log('Map style not loaded yet');
-			}
-		}
-	}, [useRealCrashData, crashDataHook?.data, crashData, heatRadius, heatIntensity, heatVisible, pointsVisible]);
 
 	useEffect(() => {
 		const el = containerRef.current;
@@ -236,16 +199,35 @@ export default function MapView({
 		const currentCrashDataHook = crashDataHook || internalCrashDataHook;
 		const activeData = crashData.length > 0 ? crashData : currentCrashDataHook.data;
 		console.log('Initializing map data, activeData length:', activeData.length);
+		
+		const initializeData = async () => {
 			if (useRealCrashData && activeData.length > 0) {
 				console.log('Using real crash data');
+				if (useAIMagnitudes) {
+					setIsLoadingAIPredictions(true);
+					console.log('🤖 Using AI-enhanced real crash data...');
+					dcDataRef.current = await convertCrashDataToGeoJSONWithAI(activeData);
+					setIsLoadingAIPredictions(false);
+				} else {
 					dcDataRef.current = convertCrashDataToGeoJSON(activeData);
+				}
 			} else if (!useRealCrashData) {
 				console.log('Using synthetic data');
+				if (useAIMagnitudes) {
+					setIsLoadingAIPredictions(true);
+					console.log('🤖 Using AI-enhanced synthetic data...');
+					dcDataRef.current = await generateDCPointsWithAI(900);
+					setIsLoadingAIPredictions(false);
+				} else {
 					dcDataRef.current = generateDCPoints(900);
+				}
 			} else {
 				console.log('No data available yet, using empty data');
 				dcDataRef.current = { type: 'FeatureCollection' as const, features: [] };
 			}
+		};
+		
+		initializeData().catch(console.error);
 
 		const computeNearbyStats = async (center: [number, number], radiusMeters = 300) => {
 			try {

web/src/app/components/UnifiedControlPanel.tsx

@@ -2,6 +2,7 @@
 
 import React, { useState, useEffect } from 'react';
 import { UseCrashDataResult } from '../hooks/useCrashData';
+import { getCircuitBreakerStatus } from '../../lib/crashMagnitudeApi';
 
 interface UnifiedControlPanelProps {
 	// Map controls props
@@ -17,6 +18,8 @@ interface UnifiedControlPanelProps {
 	onChangeIntensity: (v: number) => void;
 	gradientRoutes: boolean;
 	onToggleGradientRoutes: (v: boolean) => void;
+	useAIMagnitudes: boolean;
+	onToggleAIMagnitudes: (v: boolean) => void;
 	
 	// Crash data controls props
 	crashDataHook: UseCrashDataResult;
@@ -36,6 +39,8 @@ export default function UnifiedControlPanel({
 	onChangeIntensity,
 	gradientRoutes,
 	onToggleGradientRoutes,
+	useAIMagnitudes,
+	onToggleAIMagnitudes,
 	crashDataHook,
 	onDataLoaded
 }: UnifiedControlPanelProps) {
@@ -59,6 +64,7 @@ export default function UnifiedControlPanel({
 	const [isMapControlsSectionOpen, setIsMapControlsSectionOpen] = useState(getInitialMapControlsState);
 	const [isCrashDataSectionOpen, setIsCrashDataSectionOpen] = useState(getInitialCrashDataState);
 	const [isHydrated, setIsHydrated] = useState(false);
+	const [aiApiStatus, setAiApiStatus] = useState<{ isOpen: boolean; failures: number }>({ isOpen: false, failures: 0 });
 	
 	// Load localStorage values after hydration
 	useEffect(() => {
@@ -77,7 +83,31 @@ export default function UnifiedControlPanel({
 		}
 		
 		setIsHydrated(true);
-	}, []);	// Crash data state
+	}, []);
+	
+	// Check AI API status when AI magnitudes are enabled
+	useEffect(() => {
+		if (useAIMagnitudes) {
+			const checkApiStatus = async () => {
+				try {
+					const status = await getCircuitBreakerStatus();
+					setAiApiStatus(status);
+				} catch (error) {
+					console.error('Error checking API status:', error);
+					setAiApiStatus({ isOpen: true, failures: 1 });
+				}
+			};
+			
+			// Check immediately
+			checkApiStatus();
+			
+			// Check every 30 seconds
+			const interval = setInterval(checkApiStatus, 30000);
+			return () => clearInterval(interval);
+		}
+	}, [useAIMagnitudes]);
+	
+	// Crash data state
 	const { data, loading, error, pagination, loadMore, refresh, yearFilter, setYearFilter } = crashDataHook;
 	const [currentYear, setCurrentYear] = useState('2024'); // Default to prevent hydration mismatch
 	const [selectedYear, setSelectedYear] = useState<string>('2024'); // Default value
@@ -236,6 +266,29 @@ export default function UnifiedControlPanel({
 									<input type="checkbox" checked={gradientRoutes} onChange={(e) => onToggleGradientRoutes(e.target.checked)} />
 								</div>
 
+								<div className="mc-row">
+									<label className="mc-label">
+										AI Magnitudes 🤖
+										<span style={{
+											fontSize: 8,
+											padding: '2px 6px',
+											borderRadius: 4,
+											marginLeft: 8,
+											backgroundColor: aiApiStatus.isOpen ? '#d4edda' : '#f8d7da',
+											color: aiApiStatus.isOpen ? '#155724' : '#721c24'
+										}}>
+											{aiApiStatus.isOpen ? 'Available' : `Unavailable (${aiApiStatus.failures} failures)`}
+										</span>
+									</label>
+									<input type="checkbox" checked={useAIMagnitudes} onChange={(e) => onToggleAIMagnitudes(e.target.checked)} />
+								</div>
+								
+								{useAIMagnitudes && (
+									<div style={{ fontSize: 10, color: 'var(--text-secondary)', marginTop: -4, marginBottom: 8, lineHeight: 1.3 }}>
+										Uses AI to predict crash severity. Falls back to traditional calculation if API unavailable.
+									</div>
+								)}
+
 								<div style={{ marginBottom: 6 }}>
 									<label style={{ display: 'block', fontSize: 12 }}>Radius: {heatRadius}</label>
 									<input className="mc-range" type="range" min={5} max={100} value={heatRadius} onChange={(e) => onChangeRadius(Number(e.target.value))} style={{ width: '100%' }} />

web/src/app/lib/mapUtils.ts

@@ -1,6 +1,7 @@
 import { CrashData } from '../api/crashes/route';
+import { getCachedCrashMagnitude, CrashMagnitudePrediction } from '../../lib/crashMagnitudeApi';
 
-export type PointFeature = GeoJSON.Feature<GeoJSON.Point, { mag: number; crashData: CrashData }>;
+export type PointFeature = GeoJSON.Feature<GeoJSON.Point, { mag: number; crashData: CrashData; aiPredicted?: boolean }>;
 
 export const haversine = (a: [number, number], b: [number, number]) => {
     const toRad = (v: number) => v * Math.PI / 180;
@@ -21,8 +22,8 @@ export const convertCrashDataToGeoJSON = (crashes: CrashData[]): GeoJSON.Feature
     console.log('Sample crash data:', crashes[0]);
     
     const features: PointFeature[] = crashes.map((crash) => {
-        // Calculate severity score based on fatalities and major injuries
+        // Calculate fallback severity score based on fatalities and major injuries
-        const severityScore = Math.max(1, 
+        const fallbackSeverityScore = Math.max(1, 
             (crash.fatalDriver + crash.fatalPedestrian + crash.fatalBicyclist) * 3 + 
             (crash.majorInjuriesDriver + crash.majorInjuriesPedestrian + crash.majorInjuriesBicyclist) * 2 +
             (crash.totalVehicles + crash.totalPedestrians + crash.totalBicycles)
@@ -35,8 +36,9 @@ export const convertCrashDataToGeoJSON = (crashes: CrashData[]): GeoJSON.Feature
                 coordinates: [crash.longitude, crash.latitude]
             },
             properties: {
-                mag: Math.min(6, severityScore), // Cap at 6 for consistent visualization
+                mag: Math.min(6, fallbackSeverityScore), // Cap at 6 for consistent visualization
-                crashData: crash
+                crashData: crash,
+                aiPredicted: false // Will be updated when AI prediction is available
             }
         };
     });
@@ -52,6 +54,80 @@ export const convertCrashDataToGeoJSON = (crashes: CrashData[]): GeoJSON.Feature
     return geoJSON;
 };
 
+/**
+ * Enhanced version that fetches AI predictions for crash magnitudes
+ */
+export const convertCrashDataToGeoJSONWithAI = async (crashes: CrashData[]): Promise<GeoJSON.FeatureCollection> => {
+    console.log('🤖 Converting crash data to GeoJSON with AI predictions:', crashes.length, 'crashes');
+    
+    // Start with the basic conversion
+    const baseGeoJSON = convertCrashDataToGeoJSON(crashes);
+    
+    // Limit concurrent API calls to avoid overwhelming the API
+    const BATCH_SIZE = 10;
+    const enhancedFeatures = [...baseGeoJSON.features];
+    let successfulPredictions = 0;
+    
+    for (let i = 0; i < crashes.length; i += BATCH_SIZE) {
+        const batch = crashes.slice(i, i + BATCH_SIZE);
+        const batchPromises = batch.map(async (crash, batchIndex) => {
+            const featureIndex = i + batchIndex;
+            
+            try {
+                // Get AI prediction for this crash location
+                const prediction = await getCachedCrashMagnitude(crash.latitude, crash.longitude);
+                
+                if (prediction && typeof prediction.prediction === 'number') {
+                    // Scale the roadcast API prediction to a reasonable range for visualization
+                    // Roadcast returns values like 47, so we'll scale them to 1-10 range
+                    let scaledMagnitude = prediction.prediction;
+                    
+                    // If the value seems to be in the roadcast range (typically 0-100), scale it down
+                    if (prediction.prediction > 20) {
+                        scaledMagnitude = Math.max(1, Math.min(10, Math.round(prediction.prediction / 10)));
+                    } else {
+                        scaledMagnitude = Math.max(1, Math.min(10, Math.round(prediction.prediction)));
+                    }
+                    
+                    console.log(`🎯 Scaled magnitude from ${prediction.prediction} to ${scaledMagnitude}`);
+                    
+                    enhancedFeatures[featureIndex] = {
+                        ...enhancedFeatures[featureIndex],
+                        properties: {
+                            ...enhancedFeatures[featureIndex].properties,
+                            mag: scaledMagnitude,
+                            aiPredicted: true
+                        }
+                    };
+                    
+                    return true; // Success
+                }
+            } catch (error) {
+                console.warn(`⚠️ Failed to get AI prediction for crash ${featureIndex}:`, error);
+            }
+            
+            return false; // Failed
+        });
+        
+        const results = await Promise.allSettled(batchPromises);
+        successfulPredictions += results.filter(r => r.status === 'fulfilled' && r.value === true).length;
+        
+        // Small delay between batches to be nice to the API
+        if (i + BATCH_SIZE < crashes.length) {
+            await new Promise(resolve => setTimeout(resolve, 100));
+        }
+    }
+    
+    const enhancedGeoJSON = {
+        type: 'FeatureCollection' as const,
+        features: enhancedFeatures as PointFeature[]
+    };
+    
+    console.log(`✅ Enhanced GeoJSON with ${successfulPredictions}/${crashes.length} AI predictions`);
+    
+    return enhancedGeoJSON;
+};
+
 export const generateDCPoints = (count = 500) => {
     const center = { lon: -77.0369, lat: 38.9072 };
     const features: PointFeature[] = [];
@@ -93,13 +169,124 @@ export const generateDCPoints = (count = 500) => {
         features.push({ 
             type: 'Feature', 
             geometry: { type: 'Point', coordinates: [lon, lat] }, 
-            properties: { mag, crashData: syntheticCrash } 
+            properties: { mag, crashData: syntheticCrash, aiPredicted: false } 
         });
     }
 
     return { type: 'FeatureCollection', features } as GeoJSON.FeatureCollection<GeoJSON.Geometry>;
 };
 
+/**
+ * Enhanced version of generateDCPoints that uses AI predictions
+ */
+export const generateDCPointsWithAI = async (count = 500) => {
+    const center = { lon: -77.0369, lat: 38.9072 };
+    const features: PointFeature[] = [];
+
+    const randNormal = () => {
+        let u = 0, v = 0;
+        while (u === 0) u = Math.random();
+        while (v === 0) v = Math.random();
+        return Math.sqrt(-2.0 * Math.log(u)) * Math.cos(2.0 * Math.PI * v);
+    };
+
+    // Generate locations first
+    const locations = [];
+    for (let i = 0; i < count; i++) {
+        const radius = Math.abs(randNormal()) * 0.02;
+        const angle = Math.random() * Math.PI * 2;
+        const lon = center.lon + Math.cos(angle) * radius;
+        const lat = center.lat + Math.sin(angle) * radius;
+        locations.push({ lon, lat, index: i });
+    }
+
+    // Get AI predictions in batches to avoid overwhelming the API
+    console.log(`🤖 Getting AI predictions for ${count} synthetic points...`);
+    const BATCH_SIZE = 20;
+    const predictions: (any | null)[] = new Array(count).fill(null);
+    let successfulPredictions = 0;
+
+    for (let i = 0; i < locations.length; i += BATCH_SIZE) {
+        const batch = locations.slice(i, i + BATCH_SIZE);
+        const batchPromises = batch.map(async (location) => {
+            try {
+                const prediction = await getCachedCrashMagnitude(location.lat, location.lon);
+                return prediction;
+            } catch (error) {
+                console.warn(`⚠️ Failed to get AI prediction for synthetic point ${location.index}:`, error);
+                return null;
+            }
+        });
+
+        const batchResults = await Promise.allSettled(batchPromises);
+        batchResults.forEach((result, batchIndex) => {
+            const globalIndex = i + batchIndex;
+            if (result.status === 'fulfilled') {
+                predictions[globalIndex] = result.value;
+                if (result.value) successfulPredictions++;
+            }
+        });
+
+        // Small delay between batches
+        if (i + BATCH_SIZE < locations.length) {
+            await new Promise(resolve => setTimeout(resolve, 50));
+        }
+    }
+
+    // Create features with AI predictions or fallback magnitudes
+    for (let i = 0; i < count; i++) {
+        const location = locations[i];
+        const prediction = predictions[i];
+        
+        // Use AI prediction if available, otherwise use random magnitude
+        let mag: number;
+        let aiPredicted = false;
+        
+        if (prediction && typeof prediction.prediction === 'number') {
+            // Scale the roadcast API prediction to a reasonable range for visualization
+            if (prediction.prediction > 20) {
+                mag = Math.max(1, Math.min(10, Math.round(prediction.prediction / 10)));
+            } else {
+                mag = Math.max(1, Math.min(10, Math.round(prediction.prediction)));
+            }
+            console.log(`🎯 Synthetic point scaled magnitude from ${prediction.prediction} to ${mag}`);
+            aiPredicted = true;
+        } else {
+            mag = Math.round(Math.max(1, Math.abs(randNormal()) * 6));
+        }
+        
+        // Create synthetic crash data for backward compatibility
+        const syntheticCrash: CrashData = {
+            id: `synthetic-${i}`,
+            latitude: location.lat,
+            longitude: location.lon,
+            reportDate: new Date().toISOString(),
+            address: `Synthetic Location ${i}`,
+            ward: 'Ward 1',
+            totalVehicles: Math.floor(Math.random() * 3) + 1,
+            totalPedestrians: Math.floor(Math.random() * 2),
+            totalBicycles: Math.floor(Math.random() * 2),
+            fatalDriver: 0,
+            fatalPedestrian: 0,
+            fatalBicyclist: 0,
+            majorInjuriesDriver: Math.floor(Math.random() * 2),
+            majorInjuriesPedestrian: 0,
+            majorInjuriesBicyclist: 0,
+            speedingInvolved: Math.floor(Math.random() * 2),
+        };
+        
+        features.push({ 
+            type: 'Feature', 
+            geometry: { type: 'Point', coordinates: [location.lon, location.lat] }, 
+            properties: { mag, crashData: syntheticCrash, aiPredicted } 
+        });
+    }
+
+    console.log(`✅ Generated ${count} synthetic points with ${successfulPredictions} AI predictions`);
+
+    return { type: 'FeatureCollection', features } as GeoJSON.FeatureCollection<GeoJSON.Geometry>;
+};
+
 // Calculate crash density along a route path
 export const calculateRouteCrashDensity = (
     routeCoordinates: [number, number][],

web/src/app/page.tsx

@@ -18,6 +18,7 @@ export default function Home() {
 	const [heatRadius, setHeatRadius] = useState(16);
 	const [heatIntensity, setHeatIntensity] = useState(1);
 	const [gradientRoutes, setGradientRoutes] = useState(true);
+	const [useAIMagnitudes, setUseAIMagnitudes] = useState(true); // Default to true since roadcast API is reliable
 
 	const [popup, setPopup] = useState<PopupData>(null);
 	const [popupVisible, setPopupVisible] = useState(false);
@@ -66,6 +67,8 @@ export default function Home() {
 					onChangeIntensity={(v) => setHeatIntensity(v)}
 					gradientRoutes={gradientRoutes}
 					onToggleGradientRoutes={(v) => setGradientRoutes(v)}
+					useAIMagnitudes={useAIMagnitudes}
+					onToggleAIMagnitudes={(v) => setUseAIMagnitudes(v)}
 					crashDataHook={crashDataHook}
 				/>
 
@@ -79,6 +82,7 @@ export default function Home() {
 					crashData={crashDataHook.data}
 					crashDataHook={crashDataHook}
 					isMapPickingMode={isMapPickingMode}
+					useAIMagnitudes={useAIMagnitudes}
 					onMapReady={(m) => { mapRef.current = m; }}
 					onPopupCreate={(p) => { setPopupVisible(false); setPopup(p); requestAnimationFrame(() => setPopupVisible(true)); }}
 				/>

web/src/lib/crashMagnitudeApi.ts

@@ -0,0 +1,255 @@
+/**
+ * API service for crash magnitude prediction using roadcast model
+ */
+
+export interface CrashMagnitudePrediction {
+  prediction: number;
+  confidence?: number;
+}
+
+export interface CrashMagnitudeRequest {
+  source: {
+    lat: number;
+    lon: number;
+  };
+  destination: {
+    lat: number;
+    lon: number;
+  };
+}
+
+export interface CrashMagnitudeResponse {
+  prediction: CrashMagnitudePrediction;
+  called_with: string;
+  diagnostics?: {
+    input_dim: number;
+  };
+  index?: number;
+}
+
+/**
+ * Get crash magnitude prediction from roadcast API
+ * Simplified version that always tries to get the prediction
+ */
+export async function getCrashMagnitudePrediction(
+  sourceLat: number,
+  sourceLon: number,
+  destLat: number,
+  destLon: number
+): Promise<CrashMagnitudePrediction | null> {
+
+  try {
+    const requestBody: CrashMagnitudeRequest = {
+      source: {
+        lat: sourceLat,
+        lon: sourceLon
+      },
+      destination: {
+        lat: destLat,
+        lon: destLon
+      }
+    };
+
+    console.log('<27> Requesting crash magnitude from roadcast API:', requestBody);
+
+    // Create fetch options with timeout
+    const fetchOptions: RequestInit = {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+        'Accept': 'application/json',
+      },
+      body: JSON.stringify(requestBody),
+    };
+
+    // Add timeout if AbortSignal.timeout is supported
+    try {
+      if (typeof AbortSignal !== 'undefined' && 'timeout' in AbortSignal) {
+        fetchOptions.signal = AbortSignal.timeout(10000); // 10 second timeout
+      }
+    } catch (e) {
+      // AbortSignal.timeout not supported, continue without timeout
+      console.log('⚠️ AbortSignal.timeout not supported, continuing without timeout');
+    }
+
+    const response = await fetch('http://localhost:5000/predict', fetchOptions);
+
+    if (!response.ok) {
+      console.error('❌ Roadcast API error:', response.status, response.statusText);
+      return null;
+    }
+
+    const data: CrashMagnitudeResponse = await response.json();
+    console.log('✅ Roadcast magnitude prediction received:', data);
+    
+    // Handle roadcast API response format
+    // The roadcast API returns the magnitude in the 'index' field
+    if (data.index !== undefined) {
+      console.log('🎯 Using roadcast index as crash magnitude:', data.index);
+      return { 
+        prediction: data.index, 
+        confidence: 0.95 // High confidence for roadcast model
+      };
+    } else if (data.prediction && typeof data.prediction === 'object' && data.prediction.prediction !== undefined) {
+      // Fallback: Response format: { prediction: { prediction: number } }
+      return data.prediction;
+    } else if (typeof data.prediction === 'number') {
+      // Fallback: Response format: { prediction: number }
+      return { prediction: data.prediction };
+    }
+
+    console.warn('⚠️ No usable magnitude data in roadcast API response:', data);
+    return null;
+
+  } catch (error) {
+    
+    if (error instanceof Error) {
+      if (error.name === 'AbortError') {
+        console.warn('⏰ Crash magnitude API request timed out');
+      } else if (error.message.includes('fetch')) {
+        console.warn('🌐 Network error accessing crash magnitude API:', error.message);
+      } else {
+        console.warn('❌ Error fetching crash magnitude prediction:', error.message);
+      }
+    } else {
+      console.warn('❌ Unknown error fetching crash magnitude prediction:', error);
+    }
+    return null;
+  }
+}
+
+/**
+ * Get crash magnitude for a single point (using same point for source and destination)
+ */
+export async function getPointCrashMagnitude(
+  lat: number,
+  lon: number
+): Promise<CrashMagnitudePrediction | null> {
+  return getCrashMagnitudePrediction(lat, lon, lat, lon);
+}
+
+/**
+ * Batch get crash magnitude predictions for multiple locations
+ */
+export async function getBatchCrashMagnitudes(
+  locations: Array<{ lat: number; lon: number; id?: string }>
+): Promise<Array<{ prediction: CrashMagnitudePrediction | null; id?: string }>> {
+  const results = await Promise.allSettled(
+    locations.map(async (location) => {
+      const prediction = await getPointCrashMagnitude(location.lat, location.lon);
+      return { prediction, id: location.id };
+    })
+  );
+
+  return results.map((result, index) => {
+    if (result.status === 'fulfilled') {
+      return result.value;
+    } else {
+      console.error(`❌ Failed to get magnitude for location ${index}:`, result.reason);
+      return { prediction: null, id: locations[index].id };
+    }
+  });
+}
+
+/**
+ * Cache for magnitude predictions to avoid repeated API calls
+ */
+const magnitudeCache = new Map<string, { prediction: CrashMagnitudePrediction; timestamp: number }>();
+const CACHE_DURATION = 5 * 60 * 1000; // 5 minutes
+
+/**
+ * Status tracking for roadcast API (simplified - always available)
+ */
+
+function isCircuitBreakerOpen(): boolean {
+  // Roadcast API is local and reliable, always return false
+  return false;
+}
+
+function recordCircuitBreakerFailure(): void {
+  // Not needed for local roadcast API, but kept for compatibility
+}
+
+function recordCircuitBreakerSuccess(): void {
+  // Not needed for local roadcast API, but kept for compatibility
+}
+
+function getCacheKey(lat: number, lon: number): string {
+  return `${lat.toFixed(6)},${lon.toFixed(6)}`;
+}
+
+/**
+ * Get cached crash magnitude or fetch if not available/expired
+ */
+export async function getCachedCrashMagnitude(
+  lat: number,
+  lon: number
+): Promise<CrashMagnitudePrediction | null> {
+  const cacheKey = getCacheKey(lat, lon);
+  const cached = magnitudeCache.get(cacheKey);
+  
+  if (cached && Date.now() - cached.timestamp < CACHE_DURATION) {
+    console.log('📦 Using cached magnitude prediction for:', cacheKey);
+    return cached.prediction;
+  }
+
+  const prediction = await getPointCrashMagnitude(lat, lon);
+  
+  if (prediction) {
+    magnitudeCache.set(cacheKey, {
+      prediction,
+      timestamp: Date.now()
+    });
+  }
+
+  return prediction;
+}
+
+/**
+ * Get current status of the roadcast API by testing connection
+ */
+export async function getCircuitBreakerStatus(): Promise<{ isOpen: boolean; failures: number; resetTime?: number }> {
+  try {
+    // Test the roadcast API with a simple request
+    const controller = new AbortController();
+    const timeoutId = setTimeout(() => controller.abort(), 3000); // 3 second timeout
+    
+    const response = await fetch('http://localhost:5000/predict', {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+      },
+      body: JSON.stringify({
+        source: { lat: 38.9, lon: -77.0 },
+        destination: { lat: 38.91, lon: -77.01 }
+      }),
+      signal: controller.signal
+    });
+    
+    clearTimeout(timeoutId);
+    
+    if (response.ok) {
+      const data = await response.json();
+      console.log('🟢 Roadcast API status check successful:', data.index);
+      return {
+        isOpen: false,  // API is available
+        failures: 0,
+        resetTime: undefined
+      };
+    } else {
+      console.log('🔴 Roadcast API status check failed:', response.status);
+      return {
+        isOpen: true,   // API returned error
+        failures: 1,
+        resetTime: undefined
+      };
+    }
+  } catch (error) {
+    console.log('🔌 Roadcast API unavailable:', error);
+    return {
+      isOpen: true,   // API is unavailable
+      failures: 1,
+      resetTime: undefined
+    };
+  }
+}