Juris AI - Vectorless Legal RAG with PageIndex

Juris AI is a legal document question-answering application built with a vectorless RAG approach using PageIndex. It extracts text from uploaded PDF legal documents, builds a lightweight page-level index, retrieves the most relevant pages for a user query, and generates grounded answers with page citations.

Features

• Upload and process PDF legal documents
• Extract page-level text using PyMuPDF
• Build a vectorless PageIndex without embeddings or a vector database
• Route user questions to the most relevant document pages
• Generate legal answers using Groq-hosted LLMs
• Cite the pages used to answer each query
• React frontend for document upload and chat
• FastAPI backend with PDF upload and chat endpoints

Tech Stack

Backend

• Python
• FastAPI
• PyMuPDF
• Groq API
• Pydantic

Frontend

• React
• Vite
• Axios
• Lucide React

Project Structure

page_index/
├── core/
│   ├── config.py
│   ├── models.py
│   └── page_index.py
├── frontend/
│   ├── public/
│   ├── src/
│   ├── package.json
│   └── vite.config.js
├── uploads/
├── main.py
├── requirements.txt
├── test_bot.py
└── README.md

How It Works

1. A user uploads a PDF legal document.
2. The backend extracts text from each page using PyMuPDF.
3. A PageIndex is generated from page-level text previews.
4. When the user asks a question, the router identifies the most relevant pages from the PageIndex.
5. The answer generator uses only the retrieved pages to produce a response.
6. The response includes page citations so the answer stays grounded in the source document.

Backend Setup

Create and activate a virtual environment:

python -m venv venv
.\venv\Scripts\activate

Install dependencies:

pip install -r requirements.txt

Create a .env file in the project root:

GROQ_API_KEY=your-groq-api-key
GROQ_MODEL=llama-3.3-70b-versatile
UPLOAD_DIR=./uploads

Run the backend:

python main.py

The backend will run at:

http://localhost:8000

Frontend Setup

Move into the frontend directory:

cd frontend

Install dependencies:

npm install

Run the development server:

npm run dev

The frontend will run at:

http://localhost:5173

API Endpoints

Upload Document

POST /upload

Uploads a PDF document and builds its PageIndex.

Chat with Document

POST /chat

Sends a user query for a previously uploaded document.

Example request:

{
  "document_id": "uploaded-document-id",
  "query": "What are the termination clauses in this agreement?"
}

Environment Variables

Variable	Description
GROQ_API_KEY	API key used to access Groq models
GROQ_MODEL	Model used for page routing and answer generation
UPLOAD_DIR	Directory where uploaded PDFs and generated indexes are stored

Notes

• Do not commit .env files or API keys.
• Do not commit uploaded legal documents unless they are public sample files.
• This project is designed for document-grounded assistance and should not be treated as legal advice.