EcoSortAI ♻️

A cross‑platform app that classifies waste using AI and helps users find nearby recycling centers.

---

🚀 Overview

EcoSortAI uses a TensorFlow/Keras CNN to classify plastic waste types from images and a Flutter app to deliver a clean, engaging experience. The FastAPI backend handles authentication, user points, leaderboards, and ML inference.

• AI waste classification: PET, HDPE, LDPE, PP, PS
• Recycling centers: View centers, materials they accept, and operating hours
• Gamification: Earn points and badges for recycling actions
• Cross‑platform: Android, iOS, Web, Windows, macOS, Linux

---

🎬 Demo Screenshots

---

🏗️ Architecture

• frontend/: Flutter app (UI, routing, services)
• backend/: FastAPI service (auth, users, ML endpoints, SQLite)
• ml/: Training, preprocessing, prediction, and exported models

Data Flow:

1. User captures or picks an image in the app
2. App sends the image to FastAPI /ml/predict
3. Backend loads/uses exported model and returns label + confidence
4. User actions can update points and badges via /points

---

🗂️ Project Structure

EcoSortAI/
├─ backend/
│  ├─ app/
│  │  ├─ auth/              # Auth routes, JWT, password hashing
│  │  ├─ users/             # User points, profile, leaderboard
│  │  ├─ ml_api/            # ML inference endpoint
│  │  ├─ database/          # SQLAlchemy models & session
│  │  └─ main.py            # FastAPI app entry
│  └─ backend/eco_sort_ai.db# SQLite database (created at runtime)
├─ frontend/
│  └─ lib/                  # Flutter code (config, screens, services)
├─ ml/
│  ├─ train.py              # Training script
│  ├─ predict.py            # Local prediction helper
│  ├─ export_model.py       # Export model for serving
│  └─ exported_model/       # Saved model used by backend
└─ Images/                  # Screenshots used in README

---

🏁 Quick Start

Prerequisites

• Python 3.8+
• pip
• Flutter SDK (latest stable)
• (Optional) CUDA for GPU acceleration

1) Backend Setup

1. Navigate to backend directory:

   cd backend

2. Create and activate a virtual environment (recommended):

   python -m venv .venv
   # Windows
   .venv\Scripts\activate
   # macOS/Linux
   source .venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. Run the server:

   uvicorn app.main:app --host 0.0.0.0 --port 8000

2) Frontend Setup

1. Navigate to frontend directory:

   cd frontend

2. Install Flutter dependencies:

   flutter pub get

3. Configure API base URL in frontend/lib/config/app_config.dart:

   class AppConfig {
     // For Android emulator
     // static const String serverUrl = "http://10.0.2.2:8000";
   
     // For localhost on desktop/iOS simulator
     // static const String serverUrl = "http://localhost:8000";
   
     // For device on same LAN
     static const String serverUrl = "http://<YOUR_LOCAL_IP>:8000";
   }

4. Run the app:

   flutter run

---

🔌 API Reference (FastAPI)

Base URL: http://<server>:8000

Auth

• POST /auth/signup

  curl -X POST http://localhost:8000/auth/signup \
    -H "Content-Type: application/json" \
    -d '{"username":"alice","email":"[email protected]","password":"secret"}'

• POST /auth/login → returns { access_token }

  TOKEN=$(curl -s -X POST http://localhost:8000/auth/login \
    -H "Content-Type: application/json" \
    -d '{"username":"alice","password":"secret"}' | jq -r .access_token)

Users

• GET /users/me (Bearer token required)

  curl -H "Authorization: Bearer $TOKEN" http://localhost:8000/users/me

• POST /points (awards points and updates badge)

  curl -X POST http://localhost:8000/points \
    -H "Authorization: Bearer $TOKEN" \
    -H "Content-Type: application/json" \
    -d '{"plastic_type":"PET"}'

• GET /leaderboard

  curl -H "Authorization: Bearer $TOKEN" http://localhost:8000/leaderboard

Machine Learning

• POST /ml/predict (multipart image upload)

  curl -X POST http://localhost:8000/ml/predict \
    -F "file=@/path/to/image.jpg"

  Response:

  {
    "label": "PET",
    "confidence": 0.94,
    "info": {}
  }

---

🤖 Machine Learning Details

• Model: TensorFlow/Keras CNN.
• Classes: PET, HDPE, LDPE, PP, PS.
• Exported model path: ml/exported_model/ (used by backend).

Train

cd ml
python train.py

Predict locally

cd ml
python predict.py --image path/to/image.jpg

Export model for serving

cd ml
python export_model.py

---

�️ Database

• Engine: SQLite
• File: backend/backend/eco_sort_ai.db (auto-created)

---

🧪 Development Tips

• Hot reload (Flutter): flutter run and edit Dart files
• Interactive docs (FastAPI): Open http://localhost:8000/docs
• Testing auth quickly: Use a REST client (Insomnia/Postman) to store the Bearer token

---

🩺 Troubleshooting

• 401 Unauthorized: Ensure you pass Authorization: Bearer <token> from /auth/login.
• Cannot reach backend: Set proper serverUrl in app_config.dart and ensure device/emulator can reach your machine’s IP.
• Model errors: Make sure ml/exported_model/ exists and matches what predict.py expects.

---

🔐 Security Notes

• JWT secret is currently in code for development. For production, load SECRET_KEY from environment variables and rotate regularly.
• Validate and size-limit uploaded files for /ml/predict in production.

---

🤝 Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss the proposal.

• Fork the repo
• Create a feature branch
• Commit your changes
• Open a PR

---

📄 License

MIT

---

👤 Authors

• Rishika
• Saurav Kumar