⚖️ GenAI Legal Assistant

A Retrieval-Augmented Generation (RAG)-based AI legal assistant built using the Indian Constitution.

This system retrieves relevant constitutional articles using hybrid retrieval techniques and generates context-aware legal responses using a local LLM.

---

Features

• Hybrid Retrieval (Dense + BM25)
• FAISS Vector Search
• Cross-Encoder Reranking
• Clause-Aware Chunking
• Intent Classification
• Guardrails for Unsafe Queries
• Local LLM Reasoning (Mistral 7B)
• Evaluation Framework
• Streamlit Web Interface

---

Tech Stack

• Python
• Streamlit
• FAISS
• Sentence Transformers
• BM25
• Cross Encoder
• Llama.cpp
• Mistral 7B

---

System Architecture

User Query ↓ Intent Classification ↓ Hybrid Retrieval (FAISS + BM25) ↓ Cross Encoder Reranking ↓ Context Injection ↓ Mistral 7B Reasoning ↓ Final Legal Response

---

Project Structure

LEGAL_AI_ASSISTANT/
│
├── app/
├── models/
├── output/
├── streamlit_app.py
├── main.py
├── requirements.txt
└── README.md

---

Installation

Clone Repository

git clone https://github.com/ayanmitra07/LEGAL_AI_ASSISTANT.git
cd LEGAL_AI_ASSISTANT

Install Dependencies

pip install -r requirements.txt

---

Run Streamlit App

streamlit run streamlit_app.py

---

Sample Queries

• What is Article 21?
• Can courts interfere in elections?
• What protections exist against arbitrary arrest?
• Can Parliament amend fundamental rights?

---

Evaluation Metrics

The project includes evaluation using:

• Recall@K
• Clause Recall@K
• Mean Reciprocal Rank (MRR)

---

Future Improvements

• FastAPI Backend
• Cloud Deployment
• Chat History
• PDF Upload
• Multi-query Retrieval
• API-based LLM Integration
• Dockerization

---

Disclaimer

This system is for educational and informational purposes only and does not constitute legal advice.