FreshLogic is an advanced Agentic AI platform designed to prevent food spoilage during agricultural transit. It combines Real-time Weather Telemetry, Ensemble ML Models, and Generative AI (Gemini 2.5) to provide farmers and logistics operators with actionable intelligence.

*Regressor converted to classification using 0.5 threshold

Training Dataset: 92,000 samples across 92 crop types, generated using Arrhenius Equation (Q10 Spoilage Physics)

• Powered by Google Gemini 2.5 Flash with function calling
• RAG System retrieves crop-specific storage rules from 92-crop knowledge base
• Generates actionable "Race Engineer" style recommendations

• Regressor: Predicts continuous spoilage risk (0.0 - 1.0)
• Classifier: Confirms Safe/Spoiled classification with confidence
• Combined: Higher confidence when both models agree

• Routing: OSRM (Open Source Routing Machine) for real driving paths
• Geocoding: Nominatim (OpenStreetMap) for city → coordinates
• Weather: OpenWeatherMap for live conditions at each waypoint

• Temperature, humidity, and risk tracked at every checkpoint
• Danger zone detection with exposure time calculation
• Cumulative risk progression throughout the route

• 3-Panel workspace (Monitor, Chat, History)
• Real-time charts with risk progression
• NLP queries supported (e.g., "Analyze mango shipment from Nashik to Mumbai")

Before starting, ensure you have:

Create a .env file in the backend/ folder:

# Required
GEMINI_API_KEY=your_gemini_api_key_here

# Optional but recommended for real weather/routing data
GOOGLE_API_KEY=your_google_cloud_api_key_here

# Optional fallback (deprecated - Google Weather API is now primary)
# OPENWEATHER_API_KEY=your_openweather_api_key_here

Get your keys from:

• Gemini: https://aistudio.google.com/apikey
• Google Cloud (Maps/Routes/Weather): https://console.cloud.google.com/apis/credentials

Note: Without GOOGLE_API_KEY, the app uses free fallbacks (OSRM for routing, simulated weather data). With the key, you get real Google Weather API, Routes API, Air Quality, and more.

cd FreshLogic

# From the project root directory
python train_model.py

This will train and save:

• model/baseline_model.pkl - Regressor
• model/classifier_model.pkl - Classifier
• model/ensemble_model.pkl - Combined Ensemble

Expected Output:

============================================================
🌾 FreshLogic Ensemble Model Training
============================================================
   Samples: 92,000
   Crops: 92

📈 Training REGRESSOR (RandomForest)...
   ✅ MAE: 0.0344
   ✅ R² Score: 0.9741

🏷️  Training CLASSIFIER (RandomForest)...
   ✅ Accuracy: 0.9593
   ✅ F1 Score: 0.9626

🔗 Creating ENSEMBLE Model...
   💾 Saved: ensemble_model.pkl

Copy-Item -Path "model\*.pkl" -Destination "backend\model\" -Force

# Navigate to backend directory
cd backend

# Create virtual environment (first time only)
python -m venv venv

# Activate virtual environment
.\venv\Scripts\activate

# Install dependencies (first time only)
pip install -r requirements.txt

# Start the FastAPI server
python.exe -m uvicorn main:app --reload --host 0.0.0.0 --port 8000

Expected Output:

✅ Ensemble model loaded (v1.0.0)
   • Regressor: MAE=0.0344
   • Classifier: F1=0.9626
Loaded Knowledge Base: 92 items.
INFO:     Uvicorn running on http://0.0.0.0:8000
INFO:     Application startup complete.

# Navigate to frontend directory
cd frontend

# Install dependencies (first time only)
npm install

# Start the Vite dev server
npm run dev

Expected Output:

  VITE v6.x.x  ready in xxx ms

  ➜  Local:   http://localhost:5173/
  ➜  Network: http://192.168.x.x:5173/

FreshLogic/
├── backend/
│   ├── main.py                 # FastAPI application entry point
│   ├── requirements.txt        # Python dependencies
│   ├── .env                    # API keys (create this!)
│   ├── agents/
│   │   └── gemini_agent.py     # Gemini 2.5 + RAG integration
│   ├── services/
│   │   ├── telemetry_service.py    # OSRM routing + OpenWeatherMap
│   │   ├── model_inference.py      # Ensemble ML model loading
│   │   └── rag_service.py          # Vector search (numpy/pickle)
│   ├── model/
│   │   ├── baseline_model.pkl      # RandomForest Regressor
│   │   ├── classifier_model.pkl    # RandomForest Classifier
│   │   └── ensemble_model.pkl      # Combined Ensemble
│   └── data/
│       ├── crop_storage_data.json      # 92 crop storage rules
│       ├── synthetic_spoilage_data.csv # 92K training samples
│       └── generate_synthetic_data.py  # Data generation script
│
├── frontend/
│   ├── src/
│   │   ├── App.tsx             # Main React component
│   │   └── components/
│   │       ├── Dashboard.tsx       # Route input & analysis
│   │       ├── LiveMonitor.tsx     # Telemetry charts
│   │       ├── IntelligencePanel.tsx # AI recommendations
│   │       └── ChatInterface.tsx   # Conversational AI
│   ├── package.json
│   └── vite.config.ts
│
├── train_model.py              # Ensemble model training script
└── README.md                   # This file

# Check if port 8000 is in use
netstat -ano | findstr :8000

# Kill the process using the port (replace PID)
taskkill /PID <PID> /F

# Verify models exist in backend/model/
dir backend\model\

# Should show:
# baseline_model.pkl
# classifier_model.pkl  
# ensemble_model.pkl

# Verify .env file exists and has correct format
type backend\.env

# Should show:
# GEMINI_API_KEY=your_key
# OPENWEATHER_API_KEY=your_key

# Clear node_modules and reinstall
cd frontend
Remove-Item -Recurse -Force node_modules
npm install

{
  "risk_analysis": {
    "spoilage_risk": 0.234,
    "days_remaining": 7.6,
    "status": "Safe",
    "classification": "Safe",
    "safe_probability": 0.876,
    "ensemble_confidence": 0.912,
    "ensemble_used": true
  },
  "route": {
    "distance_km": 148.5,
    "duration_hours": 3.2
  },
  "waypoint_predictions": [
    {
      "waypoint_num": 1,
      "temperature": 28.5,
      "humidity": 65,
      "instant_risk": 0.12,
      "cumulative_risk": 0.05,
      "classification": "Safe"
    }
  ],
  "route_risk_analysis": {
    "temp_variance": 8.5,
    "danger_zone_count": 2,
    "danger_hours": 1.5
  }
}

• ✅ Real ML Model - Not simulated, trained on 92K physics-based samples
• ✅ Real Weather Data - Live OpenWeatherMap integration
• ✅ Real Routing - OSRM calculates actual driving paths
• ✅ Ensemble Approach - Two models combined for higher confidence
• ✅ Per-Waypoint Analysis - Risk tracked at every checkpoint
• ✅ Agentic AI - Gemini 2.5 with RAG for crop-specific advice

MIT License - Built for ML Mumbai GenAI Hackathon 2025

Made with 🌾 for Indian Farmers