AI PDF Chatbot (RAG-based)
A PDF question-answering app based on Retrieval-Augmented Generation. It parses uploaded PDFs, stores embedded chunks in a vector database and uses Google Gemini to answer questions about the documents.
- Period
- Jun 2025 Present
- Role
- Sole developer
- Status
- in progress
- Category
- ai
The problem
Reading long PDFs to find specific answers is slow; answers need to stay grounded in the document.
What I built
- Upload and parsing of PDFs, including OCR for scanned documents
- Embedding of document chunks into ChromaDB with HuggingFace embeddings
- Querying through Google Gemini with a model-selection UI and configurable temperature
- Automatic model fallback and retry with exponential backoff when rate-limited
- Microservice split: Streamlit client and FastAPI server
- Shell scripts for start, stop, status and validation, with macOS auto-start and crash recovery
Highlights
- RAG pipeline over PDFs with OCR
- Gemini models with fallback and retry on rate limits
- FastAPI backend and Streamlit client