Traffic Rules Assistant (Tamil Nadu RAG Chatbot)

An AI-powered assistant built using Retrieval-Augmented Generation (RAG) that provides context-aware answers about Tamil Nadu traffic rules, penalties, and driver rights. It combines semantic search with LLMs for accurate, explainable, and trustworthy responses.

---

Repo Structure

traffic_rules_assistant/
├── .venv/          
├── data/
|   ├── processed/
|   |    ├── faiss_index.idx
|   |    ├── TN_traffic_rules.json
|   |    └── TN_traffic_rules.txt
|   └──  TN Traffic rules.pdf
├── src/
│   ├── chunking.py 
│   ├── embedding.py 
│   ├── generator.py
│   ├── retriever.py
│   ├── main.py
│   └── text_extraction.py 
├── api/
│   └──  app.py
├── frontend/ 
│   └── index.html  
├── .gitignore
├── .python-version
├── README.md
├── requirements.txt
└── pyproject.toml

Overview

This project extracts legal content from a government-issued traffic PDF and makes it queryable via natural language questions using a custom-built RAG pipeline. It uses:

• sentence-transformers for generating document embeddings
• FAISS for fast semantic retrieval
• LangChain with Groq LLM backend for fast and accurate response generation
• FastAPI to expose the system as an API

The assistant is designed to provide concise, legally grounded, and verifiable answers.

---

Target Audience

• Citizens of Tamil Nadu
• Driving school instructors and trainees
• Traffic law educators
• AI/ML enthusiasts learning about RAG pipelines

---

Prerequisites

• Python 3.11+
• Basic terminal/CLI knowledge
• Internet access for LLM API (Groq)
• Groq API key (free tier available)

---

Installation

# Clone the repository
https://github.com/madhans476/traffic_rules_assistant.git

cd traffic_rules_assistant

# Create environment
uv venv .venv
source .venv/bin/activate  # or .venv\Scripts\activate on Windows

# Install dependencies
uv pip install -r requirements.txt

---

Environment Setup

.env file (create at root):

GROQ_API_KEY=your-groq-api-key-here

Required files:

• data/TN Traffic rules.pdf

---

Usage

1. Text Extraction

python src/text_extraction.py

2. Chunking

python src/chunking.py

3. Embedding + FAISS Index Creation

python src/embedding.py

4. CLI Chatbot

python src/main.py

5. API (FastAPI)

uvicorn api.app:app --reload
# Visit http://localhost:8000/docs

---

Data Requirements

• Input: A traffic rulebook in .pdf

• Output:

  • .txt file with full extracted text
  • .json with clean overlapping chunks
  • .idx FAISS index file

---

Testing

Manual test cases:

• Ask questions from the CLI or Swagger UI
• Evaluate LLM responses vs. original PDF

(Automated tests can be added using pytest.)

---

Configuration

Chunking:

• chunk_size = 300
• overlap = 50

FAISS:

• IndexFlatIP used for vector similarity search

LLM (Groq):

• Model: mixtral-8x7b-32768 (default)
• Temperature: 0.3

---

Methodology

1. Parse legal traffic PDF to text

2. Split text into semantic chunks

3. Embed chunks using sentence-transformers

4. Store in FAISS for fast vector search

5. On query:

   • Embed the question
   • Retrieve top-k relevant chunks
   • Pass to Groq LLM using LangChain

6. Return clean, structured answer

---

Performance

• Fast local retrieval (~50–100 ms)
• Groq LLM response: ~100 tokens/ms
• Accurate context-based answers
• No hallucinations (guardrails in prompt)

---

License

MIT License. See LICENSE file.

---

Contributing

Pull requests welcome!

• Fork the repo
• Create a new branch
• Submit a pull request with a meaningful message

---

Changelog

v1.0.0 (June 2025)

• Initial public release
• Added full pipeline: Extraction, Chunking, Embedding, Retrieval, Generation
• FastAPI API + CLI support
• Groq + LangChain integration

---

Citation

If you use this in academic work:

@misc{trafficassistant2025,
  title={Tamil Nadu Traffic Rules Assistant using RAG},
  author={Madhan S},
  year={2025},
  howpublished={\url{https://github.com/madhans476/traffic_rules_assistant}}
}

---

Contact

Maintainer: Madhan S

• Email: mail
• GitHub: @madhans476
• LinkedIn @madhans17