Lizzy AI: Legal Contract Q&A Chatbot with RAG System

This repository implements a Retriever-Augmented Generation (RAG) system for a legal contract Q&A chatbot, built by Lizzy AI. The system allows users to ask questions about legal contracts and receive informative answers.

Project Goals

• Develop a robust RAG system for legal contract Q&A.
• Explore and evaluate different chunking strategies.
• Implement techniques to improve answer accuracy and relevance.

Features

• Interactive chat interface for user queries.
• Knowledge base of legal contracts in PDF format.
• Text extraction and chunking from PDFs (simple and cosine similarity based).
• Retriever component using vector embeddings for efficient information retrieval.
• Generator component leveraging OpenAI's GPT-4 for answer generation.
• Prompt-based approach with context from retrieved contract sections.
• Evaluation framework to assess answer quality using various metrics.

Getting Started

This project requires Python libraries like Flask, PyPDF2, Langchain, OpenAI API, ragas, and a compatible GPT-4 API access.

1. Clone the repository:

git clone https://github.com/your-username/lizzy-ai-legal-qa.git

2. Install dependencies:

cd backend
pip install -r requirements.txt

3. Configure API Keys:

• Create an account with OpenAI and obtain an API key.
• (Optional) Set up access to a compatible GPT-4 API.

4. Run the front end application:

cd frontend
npm install
npm start

This will start the user interface at http://localhost:3000/ in your web browser.

4. Run the bakend end application:

python app.py

Usage

The chat interface allows users to type questions about legal contracts from the knowledge base. The system retrieves relevant contract sections and utilizes GPT-4 to generate informative answers based on the retrieved context.

Evaluation

The project includes an evaluation script that calculates metrics like answer relevancy, faithfulness, context recall, and precision. You can modify the script to use your own evaluation datasets.

Further Development

This project provides a foundation for a legal contract Q&A chatbot. Here are some potential areas for further development:

• Integration with a more advanced user interface framework.
• Exploration of alternative chunking algorithms.
• Implementation of multi-query retrieval with reranking for improved accuracy.
• Training a custom LLM specifically for legal contracts.
• Expanding the knowledge base with a wider variety of legal documents.

License

This project is licensed under the MIT License. See the LICENSE file for details.