📄 PDF Question Answering System (RAG-based AI App)

🏠 Home Screen

💬 Question Answering

🚀 Approach

Overview

The system follows a RAG (Retrieval-Augmented Generation) approach:

• Users upload a PDF document
• The backend extracts and chunks text
• Each chunk is converted into embeddings
• Embeddings are stored in a vector database
• User questions are embedded and matched against stored chunks
• Relevant context is sent to the LLM
• The LLM answers only using retrieved context

This ensures accuracy, traceability, and minimal hallucination.

Tools & Design Choices

Tool	Purpose	Reason
FastAPI	Backend API	Fast, async, production-ready
React (Vite)	Frontend	Lightweight, modern UI
ChromaDB	Vector database	Local, persistent, ideal for RAG
LangChain	RAG utilities	Clean abstraction for loaders & chunking
Sentence Transformers	Embeddings	Efficient semantic representation
OpenAI	LLM	High-quality text generation

Design Decisions:

• Used RAG to prevent hallucinations
• Chose ChromaDB for simplicity and local persistence
• Used environment variables for API key security
• Clean separation between retrieval and generation
• Minimal but elegant frontend for clarity and usability

🔧 Project Structure

pdf-qa/
├── backend/
│   ├── main.py
│   ├── rag.py
│   ├── requirements.txt
│   └── .env
│
├── frontend/
│   ├── src/
│   │   ├── App.jsx
│   │   ├── App.css
│   │   └── index.css
│   └── package.json
│
├── screenshots/
│   ├── home.png
│   └── answer.png
│
├── .gitignore
└── README.md

📦 Installation

1️⃣ Clone the repository:

git clone https://github.com/pabitrajana007/pdf-qa.git
cd pdf-qa

2️⃣ Backend Setup:

cd backend
python -m venv venv
source venv/Scripts/activate   # Windows (Git Bash)

2. Install dependencies:

pip install -r requirements.txt

3. Create .env file:

OPENAI_API_KEY=your_openai_api_key_here

4. Run backend:

python -m uvicorn main:app --reload

3️⃣ Frontend Setup:

cd frontend
npm install
npm run dev