Retrieval Augmented QA System

A local Document QA system that allows users to upload PDF documents and ask questions about their content using a Retrieval Augmented Generation (RAG) approach. This project uses a lightweight LLM to perform inference directly on your machine.

Features

• PDF Document Upload: Upload PDF files to be processed for question answering.
• Context-Aware QA: Ask questions specifically about the uploaded document.
• Local Inference: Uses LiquidAI/LFM2-1.2B-RAG model running locally (default: CPU).
• FastAPI Backend: robust and fast API for handling requests.
• Simple Web Interface: Includes a basic static frontend for testing.

Tech Stack

• Language: Python 3.13+
• Framework: FastAPI
• ML Libraries: PyTorch, Transformers, Accelerate
• PDF Processing: pypdf
• Server: Uvicorn

Prerequisites

• Python 3.13 or higher
• uv (recommended) or pip

Installation

1. Clone the repository:

   git clone <repository-url>
   cd document-qa-system

2. Install dependencies using uv:

   uv sync

   Or using pip:

   pip install .

Usage

Starting the Server

Run the application using the main entry point:

python main.py

The server will start at http://0.0.0.0:8000.

Web Interface

Open your browser and navigate to: http://localhost:8000

API Endpoints

• POST /api/upload

  • Upload a PDF file.
  • Returns a doc_id.

• POST /api/ask

  • Body: {"doc_id": "string", "question": "string"}
  • Returns the answer generated by the model.

Configuration

Configuration constants are located in app/constants/config.py:

• MODEL_NAME: Defaults to LiquidAI/LFM2-1.2B-RAG
• DEVICE: Defaults to cpu (can be changed to cuda if available)

Project Structure

document-qa-system/
├── app/
│   ├── constants/       # Configuration and constants
│   ├── static/          # Frontend static files
│   ├── inference.py     # ML model and QA logic
│   └── main.py          # FastAPI application definition
├── data/                # Example data
├── main.py              # Application entry point
├── pyproject.toml       # Project dependencies
└── README.md            # Project documentation