RAG PDF Chatbot

A full-stack Retrieval-Augmented Generation (RAG) chatbot that allows users to upload PDF documents and ask questions about their content using LLaMA/Mistral LLM.

Tech Stack

Backend

• Framework: FastAPI
• LLM: LLaMA/Mistral (via Ollama)
• Vector Database: FAISS
• Embeddings: HuggingFace sentence-transformers
• PDF Processing: PyPDF

Frontend

• Framework: React + Vite
• Styling: Custom CSS
• API Integration: Fetch API

Features

✅ PDF upload and processing
✅ Text extraction and intelligent chunking
✅ Semantic search using FAISS vector database
✅ Context-aware responses using LLaMA
✅ Professional chat interface
✅ Real-time message display

Setup Instructions

Prerequisites

1. Python 3.9+
2. Node.js 18+
3. Ollama - Install from ollama.com

Backend Setup

# Navigate to backend directory
cd backend

# Create virtual environment
python -m venv venv

# Activate virtual environment
# On Windows:
venv\Scripts\activate
# On Mac/Linux:
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Create necessary __init__.py files
touch app/__init__.py
touch app/api/__init__.py
touch app/rag/__init__.py
touch app/schemas/__init__.py
touch app/utils/__init__.py
touch app/core/__init__.py

# Create data directories
mkdir -p app/data/pdfs
mkdir -p app/data/faiss_index

Install and Run Ollama

# Install Ollama (follow instructions at ollama.com)

# Pull Mistral model
ollama pull mistral

# Keep Ollama running (in a separate terminal)
ollama run mistral

Run Backend

# From backend directory
uvicorn app.main:app --reload

# Backend will run on http://localhost:8000
# Swagger docs available at http://localhost:8000/docs

Frontend Setup

# Navigate to frontend directory
cd frontend

# Install dependencies
npm install

# Run development server
npm run dev

# Frontend will run on http://localhost:5173

Usage

1. Start Ollama: Run ollama run mistral in a terminal
2. Start Backend: Run uvicorn app.main:app --reload in backend directory
3. Start Frontend: Run npm run dev in frontend directory
4. Open Browser: Navigate to http://localhost:5173
5. Upload PDF: Click "Upload PDF" and select a PDF file
6. Ask Questions: Type questions about your PDF in the chat interface

RAG Pipeline Flow

1. PDF Upload
   ↓
2. Text Extraction (PyPDF)
   ↓
3. Text Chunking (with overlap)
   ↓
4. Generate Embeddings (HuggingFace)
   ↓
5. Store in FAISS Vector DB
   ↓
6. User Query
   ↓
7. Query Embedding
   ↓
8. Similarity Search (FAISS)
   ↓
9. Retrieve Context
   ↓
10. Generate Prompt
    ↓
11. LLM Generation (LLaMA/Mistral)
    ↓
12. Return Answer

API Endpoints

Health Check

• GET / - Check if backend is running

Upload PDF

• POST /api/upload
• Body: multipart/form-data with PDF file
• Response: Confirmation with number of chunks created

Chat

• POST /api/chat
• Body: {"question": "your question here"}
• Response: {"answer": "generated answer"}

Project Structure

rag-pdf-chatbot/
├── backend/
│   ├── app/
│   │   ├── api/          # API routes
│   │   ├── rag/          # RAG pipeline logic
│   │   ├── schemas/      # Pydantic models
│   │   ├── utils/        # Helper functions
│   │   ├── data/         # PDF storage & FAISS index
│   │   └── main.py       # FastAPI app
│   └── requirements.txt
│
└── frontend/
    ├── src/
    │   ├── components/   # React components
    │   ├── pages/        # Page components
    │   ├── services/     # API calls
    │   └── styles/       # CSS files
    └── package.json

Interview Talking Points

1. RAG Architecture: "I implemented a RAG system that combines retrieval from a vector database with generative AI to provide accurate, context-grounded answers."

2. Vector Search: "I used FAISS for efficient similarity search and HuggingFace embeddings for semantic understanding."

3. Chunking Strategy: "I implemented recursive chunking with overlap to preserve context across chunk boundaries."

4. LLM Integration: "I integrated an open-source LLaMA model via Ollama to keep the solution free and locally deployable."

5. Full-Stack: "I built a complete full-stack application with FastAPI backend and React frontend, ensuring clean separation of concerns."

6. Production Practices: "The code follows production best practices with proper error handling, schema validation, and CORS configuration."