RAG Document Chatbot
End-to-end AI system that answers questions from any PDF document in seconds using Retrieval-Augmented Generation — deployed live on Streamlit Cloud.
RAG
Architecture
6
Pipeline Steps
1,000
Chunk Size (chars)
Free
Deployment Cost
The Problem
Knowledge workers spend hours manually searching through dense PDF reports, legal documents, and research papers. Ctrl+F finds keywords but misses context. Teams needed a way to ask natural-language questions and get accurate, cited answers — instantly.
The Solution
Built a 6-step RAG pipeline: PyPDF loads the document and splits it into 1,000-character chunks with 200-character overlap. Google Generative AI Embeddings convert each chunk into vectors stored in ChromaDB. At query time, the top-k relevant chunks are retrieved and passed to gemini-1.5-flash via a ConversationalRetrievalChain — answers are strictly grounded in the retrieved context with multi-turn chat history maintained. Deployed to Streamlit Cloud with auto-deploy on GitHub push and API keys managed via Streamlit Secrets.
Results & Metrics
- Reduces document research time by 80%+ versus manual search
- Answers strictly grounded in the source document — zero hallucination
- 1,000-character chunks with 200-character overlap preserve full context
- Multi-turn conversation with chat history maintained across questions
- Deployed live on Streamlit Cloud — free hosting, auto-deploy on GitHub push