🚗 AutoBot – Intelligent Chatbot for Auto Sales

AutoBot is an AI-powered chatbot built using PyTorch, FastAPI, and MongoDB that helps users search and filter vehicles intelligently. It can handle queries related to brands, models, fuel types, price ranges, and more, and supports feedback collection for continuous improvement via scheduled model retraining.

---

🚀 Features

• 🔍 Natural Language Understanding for vehicle queries (e.g., “Show me Toyota Vitz under 5 million”)
• 📊 Price, model, brand, and type filters with fuzzy matching
• 💬 Interactive Chat Mode or REST API
• 🧠 ML Model Training with intent classification using PyTorch
• ♻️ Daily Retraining from Feedback (admin-reviewed)
• 📈 Feedback Logging with thumbs up/down reactions
• 📂 MongoDB Integration for vehicle data and feedback history

---

🧰 Tech Stack

• Python 3.12+
• PyTorch
• FastAPI
• MongoDB
• Uvicorn (for dev server)
• scikit-learn (metrics)
• matplotlib (for training metrics)

---

📁 Folder Structure

auto_sales_bot/
│
├── data/
│   └── intents.json              # Training intents
├── models/
│   └── model.pth                 # Saved PyTorch model
├── utils/
│   └── preprocessing.py          # Preprocessing utils
├── api.py                        # FastAPI endpoint
├── chatbot.py                    # Chat logic & ML inference
├── train.py                      # Training script
├── retrain_daily.py              # Scheduled retraining
├── mongo_service.py              # MongoDB interface
└── requirements.txt              # Dependencies

---

⚙️ Setup Instructions

1. Clone the repository

git clone https://github.com/Lprabodha/auto-sales-chatbot
cd auto_sales_bot

git clone https://github.com/Lprabodha/auto-sales-chatbot-widget

2. Create & activate a virtual environment

python -m venv .venv
source .venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Start MongoDB

Ensure MongoDB is running locally at mongodb://localhost:27017. Use Docker or install MongoDB natively.

Sample Data & Retraining

🚗 Sample vehicle listings JSON: data/sample_vehicles.json

Import Sample Data to MongoDB

mongoimport --uri "mongodb://localhost:27017" \
--db auto_sales_bot \
--collection vehicles \
--file data/sample_vehicles.json \
--jsonArray

5. Configure your .env file

MONGO_URI=mongodb://localhost:27017
DB_NAME=auto_sales_bot
MODEL_PATH=models/model.pth
SECRET_KEY=your-secret-key

6. Train the model (initial run)

python train.py

7. Start the API

uvicorn api:app --reload

---

📬 API Endpoints

POST /chat

Request:

{ "query": "Show me Toyota Vitz under 5 million" }

Response:

{
  "response": "Here are some Toyota Vitz available:",
  "prob": 0.97,
  "intent": "ask_brand_model",
  "suggestions": [
    {
      "id": 1,
      "model_name": "Vitz 2019",
      "vehicle_name": "Toyota Vitz",
      "year": 2019,
      "price": 9500000,
      "mileage": 60000
    }
  ]
}

POST /feedback

Request:

{
  "query": "Any Toyota cars under 4 million?",
  "response": "Here are some Toyota available:",
  "predicted_intent": "ask_price_range",
  "prob": 0.91,
  "thumbs_up": false
}

POST /retrain-now

Forces model retraining based on feedback data (after admin update).

---

⏰ Automating Daily Retraining

Use a cron job or scheduler to run the retrain script:

0 0 * * * cd /path/to/project && .venv/bin/python retrain_daily.py

This will:

• Update intents.json from admin-approved feedback
• Retrain the model if required
• Save the updated model

---

✨ Contributions

Feel free to open issues or PRs for new features, model improvements, or bug fixes.

---

👨‍💻 Authors

• Dewmina Udayashan
• Lahiru Prabodha

---

Happy coding! 😊

---

📜 License

MIT License. See LICENSE file for details.