CarbonSense AI

AI-Powered Carbon Emission & AQI Intelligence Platform for Bengaluru

---

Architecture

Frontend (React + Vite + Tailwind)
    └── Zustand Global State
        └── Map Tab (Source of Truth for Location)
            └── Predict Tab → Insights + Solutions
            └── Simulate Tab
            └── Forecast Tab
                └── Backend API (Flask)
                    └── predict_final.py (YOUR TRAINED ML MODEL)

---

Setup

1. Place your model

Copy predict_final.py and all model files into ./model/ directory:

model/
  predict_final.py
  *.pkl / *.joblib / *.h5  (your trained model files)

2. Backend

cd backend
pip install -r requirements.txt

# Create .env
cp .env.example .env
# Edit PREDICT_DIR to point to your model directory

python app.py

3. Frontend

# Root directory
cp .env.example .env
# Set VITE_BACKEND_URL=http://localhost:5000

npm install
npm run dev

4. Docker (Production)

# Build frontend first
npm run build

# Run with Docker Compose
docker-compose up --build

---

API Endpoints

Method	Endpoint	Description
POST	/api/predict	Run ML prediction
POST	/api/forecast	Generate 6-12 hour forecast
POST	/api/simulate	Simulate parameter changes
GET	/health	Health check

POST /api/predict

{
  "latitude": 12.9716,
  "longitude": 77.5946,
  "date": "2024-01-15",
  "time": "14:30"
}

POST /api/forecast

{
  "latitude": 12.9716,
  "longitude": 77.5946,
  "date": "2024-01-15",
  "hours": 12
}

POST /api/simulate

{
  "latitude": 12.9716,
  "longitude": 77.5946,
  "date": "2024-01-15",
  "time": "14:30",
  "ndvi_override": 0.6,
  "temperature_override": 25.0,
  "building_density_override": 30.0
}

---

predict_final.py Integration

The bridge (backend/services/model_bridge.py) tries multiple function signatures:

# It will try these in order:
predict(lat, lng, date, time)
predict(latitude=lat, longitude=lng, date=date, time=time)
predict({'latitude': lat, 'longitude': lng, 'date': date, 'time': time})

The function name is auto-detected from: predict, run_prediction, predict_aqi, main, get_prediction

Return value is normalized — supports both dict and object returns.

---

Tabs

Tab	Data Source
Map	OpenStreetMap + Nominatim geocoding
Predict	predict_final.py via /api/predict
Insights	Prediction result (NO extra API calls)
Simulate	/api/simulate (reruns ML with overrides)
Forecast	/api/forecast (reruns ML per hour)
Solutions	Rule-based AI engine (frontend only)