RAG Chatbot — Custom Domain ChatGPT

A production-ready Retrieval-Augmented Generation (RAG) system that lets you chat with your own PDF documents using AI.

What It Does

Upload any PDF → Ask questions in natural language → Get grounded answers with source citations. No hallucination. No guessing.

Tech Stack

Component	Technology
LLM	Google Gemini 2.5 Flash (free tier)
Embeddings	Gemini Embedding-001
Vector Store	FAISS
Keyword Search	BM25
Framework	LangChain 0.3.x
API Server	FastAPI
Language	Python 3.13

Architecture

PDF Documents ↓ DocumentIngester (PyPDF + chunking) ↓ VectorStoreManager (Gemini embeddings → FAISS index) ↓ HybridRetriever (BM25 40% + FAISS 60% → RRF merge) ↓ RAGChain (memory + condensation + Gemini → grounded answer) ↓ FastAPI Server (REST endpoints)

Key Features

• Hybrid Search — BM25 keyword + FAISS semantic via custom RRF
• Multi-turn Memory — remembers last 5 conversation turns
• Question Condensation — rewrites follow-ups as standalone questions
• Hallucination Prevention — LLM constrained to document context only
• Source Citations — every answer cites filename and page number
• Incremental Indexing — add documents without full rebuild
• REST API — FastAPI server with upload, chat, and memory endpoints

Setup

1. Clone and create virtual environment

git clone https://github.com/YOUR_USERNAME/rag_chatbot.git
cd rag_chatbot
python -m venv venv
# Windows:
venv\Scripts\activate
# Mac/Linux:
source venv/bin/activate

2. Install dependencies

pip install -r requirements.txt

3. Set up API key

Create a .env file in the root folder: GOOGLE_API_KEY=your_gemini_api_key_here GEMINI_MODEL=gemini-2.5-flash Get a free key at: https://aistudio.google.com/api-keys

4. Add your documents

Place PDF or TXT files in data/sample_docs/

5. Build the vector index

python backend/vectorstore.py

6. Start chatting

Terminal chat:

python backend/chat_cli.py

API server:

uvicorn backend.server:app --reload --port 8000

Then open http://localhost:8000/docs for interactive API docs.

API Endpoints

Method	Endpoint	Description
GET	/	Health check
GET	/status	Index and document info
POST	/chat	Ask a question
POST	/upload	Upload new document
DELETE	/memory	Clear conversation history

Project Structure

rag_chatbot/ ├── backend/ │ ├── ingester.py # PDF loading and chunking │ ├── vectorstore.py # FAISS index management │ ├── retriever.py # BM25 + FAISS hybrid search │ ├── rag_chain.py # RAG pipeline + memory │ ├── server.py # FastAPI REST server │ ├── chat_cli.py # Interactive terminal chat │ └── create_sample_pdf.py # Sample data generator ├── data/sample_docs/ # Your documents go here ├── faiss_index/ # Auto-generated vector index ├── .env # API keys (never commit) ├── .gitignore ├── requirements.txt └── README.md

Resume Keywords

RAG, LangChain, FAISS, BM25, Hybrid Search, Reciprocal Rank Fusion, Vector Embeddings, Semantic Search, FastAPI, Gemini API, NLP, LLM, Retrieval-Augmented Generation, Python, REST API