⚖️ LegalAI — Pakistani Legal Research & Study Assistant

An AI-powered, Retrieval-Augmented Generation (RAG) assistant built to make Pakistani jurisprudence searchable, understandable, and testable.

Live Demo | Report a Bug

---

📖 Overview

Navigating statutory frameworks and case law is traditionally bottlenecked by the manual review of massive, dense PDF texts. LegalAI transforms static legal documents into an interactive knowledge base.

Designed for Pakistani law students, legal researchers, and citizens, the platform allows users to upload local legal documents, query them via natural language, and generate highly targeted multiple-choice revision quizzes based strictly on the ingested material.

2026-05-20.23-24-52_1.mp4

✨ Key Features

• 📄 Dynamic Document Ingestion: Upload any legal PDF; the backend automatically chunks, embeds, and indexes the text into a localized Vector Store.
• 💬 Persona-Driven RAG Chat: Ask questions and receive answers grounded exclusively in your uploaded texts. Features three distinct response personas:
  • Student Mode: Explains concepts with an emphasis on core principles and memorization.
  • Lawyer Mode: Generates formal, strictly cited, professional interpretations.
  • Citizen Mode: Translates dense legalese into plain, accessible language.
• 🧠 Automated Quiz Synthesis: Instantly spin up N-question multiple-choice tests from the uploaded PDFs to test comprehension, complete with automated grading and legal source citations.
• 🛡️ Production-Ready Guardrails: Engineered with strict CORS policies, IP-based sliding-window rate limiting (429), and internal API header verification.

---

🏛️ System Architecture

The application follows a clean, decoupled Client-Server Architecture:

  ┌─────────────────┐             ┌───────────────────┐
  │ React UI        │ ──(HTTPS)──►│ FastAPI Backend   │
  │ (Vercel)        │◄──(REST)────│ (Railway Container)│
  └─────────────────┘             └─────────┬─────────┘
                                            │
                                    (LangChain / Embeddings)
                                            │
                                            ▼
                                  ┌───────────────────┐
                                  │ Local Vector DB   │
                                  └─────────┬─────────┘
                                            │ (Relevant Chunks)
                                            ▼
                                  ┌───────────────────┐
                                  │  Groq Cloud LPU   │
                                  │  (Mixtral/Llama3) │
                                  └───────────────────┘

Tech Stack Matrix

Layer	Technology	Purpose
Frontend	React, Vite	Responsive Single Page Application (SPA)
Backend	FastAPI, Python 3.11+	Async API routing, validation, file management
AI / NLP	Groq API, LangChain	High-speed LLM inference & Vector similarity search
Persistence	Custom Vector Store	In-memory/Disk document embedding storage
Deployment	Vercel & Railway	Edge-rendered client + Containerized Python host

---

🚀 Quickstart (Local Development)

Prerequisites

• Python 3.10+
• Node.js 18+
• A valid Groq API Key

1. Clone the repository

git clone [https://github.com/Zimal-Fatemah/LegalAI.git](https://github.com/Zimal-Fatemah/LegalAI.git)
cd LegalAI

2. Backend Setup

Open a terminal and navigate to your backend directory:

# Create and activate virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows use: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Create environment file
touch .env

Inside .env, declare the following:

GROQ_API_KEY="gsk_your_actual_api_key_here"
INTERNAL_API_KEY="your_custom_secret_passphrase"
ALLOWED_ORIGINS="http://localhost:5173,http://localhost:3000"

Boot the server:

uvicorn main:app --reload --port 8000

3. Frontend Setup

Open a second terminal window:

cd frontend # (or your specific React folder name)
npm install

Create a .env.local file in the frontend root:

VITE_API_BASE_URL="http://localhost:8000"
VITE_INTERNAL_API_KEY="your_custom_secret_passphrase"

Start the development server:

npm run dev

Visit http://localhost:5173 in your browser.

---

⚡ API Endpoint Reference

The backend exposes an async REST API. (Requires the X-API-Key header for secure routes):

Method	Endpoint	Description
GET	/api/health	Diagnostic check; returns RAG and Groq subsystem readiness
POST	/api/upload	Accepts a .pdf file (multipart/form-data) and re-indexes the Vector Store
GET	/api/documents	Returns a list of currently indexed legal PDF documents
POST	/api/chat	Accepts a prompt + Persona; returns an LLM response and cited source excerpts
POST	/api/quiz	Generates a structured JSON multiple-choice quiz based on active documents

---

🔮 Roadmap

• Urdu Language Support: Native language ingestion and LLM output parsing.
• Cloud Vector DB Migration: Move from local vector storage to persistent Pinecone / Qdrant.
• Citation Highlighting: Auto-scroll the frontend PDF viewer to the exact page referenced by the AI.

---

📄 License

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