RAG Chatbot

A production-grade Retrieval-Augmented Generation (RAG) chatbot built with LangChain, FAISS, Groq, and Streamlit. Upload documents or scrape websites and chat with your knowledge base using free LLMs.

Features

• Document Loaders — PDF (PyPDF2), DOCX, TXT, Markdown
• Smart Chunking — Recursive, Character, Token-based strategies
• Free Embeddings — sentence-transformers/all-MiniLM-L6-v2 (runs locally)
• Hybrid Search — FAISS dense + BM25 sparse with Reciprocal Rank Fusion
• Free LLMs — Groq (llama-3.1, mixtral, gemma) + HuggingFace
• RAG Evaluation — RAGAS framework (Faithfulness, Answer Relevancy, Context Precision)
• Web Scraping — BeautifulSoup + Selenium fallback
• Multilingual — Auto-detect + translate (13 languages)
• Export Chat — TXT, CSV, JSON, PDF
• Conversation Memory — Sliding window history with standalone query rewriting

Quick Start

1. Clone the repository

git clone https://github.com/ronitgulia/RAG-Chatbot
cd RAG-Chatbot

2. Install dependencies

pip install -r requirements.txt

3. Get a free Groq API key

Sign up at console.groq.com — it's completely free.

4. Run the app

streamlit run app.py

5. Usage

1. Paste your Groq API key in the sidebar → click Connect LLM
2. Go to Documents tab → upload files or scrape URLs
3. Go to Chat tab → ask questions
4. View Evaluation tab for quality metrics
5. Export your conversation in any format

Project Structure

RagChatBot/
├── app.py                  # Streamlit UI (Chat, Documents, Evaluation, Export)
├── rag_pipeline.py         # Core orchestrator
├── document_loader.py      # PDF, DOCX, TXT loaders
├── text_chunker.py         # LangChain text splitters
├── embeddings.py           # Sentence-transformers wrapper
├── vector_store.py         # FAISS + BM25 hybrid store
├── llm_provider.py         # Groq + HuggingFace providers
├── evaluation.py           # RAGAS evaluation pipeline
├── web_scraper.py          # Web scraping module
├── multilingual.py         # Language detection & translation
├── export_utils.py         # Chat export utilities
├── styles.py               # Custom Streamlit CSS
├── config.py               # Centralized configuration
├── requirements.txt        # Dependencies
└── .env.example            # Environment variable template

Evaluation Metrics

Metric	Description
Faithfulness	Is the answer grounded in retrieved context?
Answer Relevancy	Does the answer address the question?
Context Precision	Are retrieved chunks relevant to the query?

Available LLM Models (Groq — Free)

Model	Context	Best For
llama-3.1-8b-instant	128k	Fast responses (default)
llama-3.3-70b-versatile	128k	Best quality
deepseek-r1-distill-llama-70b	128k	Reasoning tasks
gemma2-9b-it	8k	Balanced
mixtral-8x7b-32768	32k	Long documents

License

MIT License