How itanishqshelar/smartrag works
Project type and purpose
• Full-stack, locally-hosted multimodal RAG (Retrieval-Augmented Generation) system with a Streamlit “chat” front-end.
• Allows users to ingest documents, images, and audio, stores metadata in SQLite, embeddings in ChromaDB/FAISS, and uses a local LLM (Llama 3.1 via Ollama) for responses.
File structure
• Top-level scripts:
• chatbot_app.py (Streamlit UI entry point)
• start.py / clean_db.py (utility scripts)
• Configuration: config.yaml + config_schema.py for Pydantic validation
• multimodal_rag/: core library
• base.py (interfaces)
• system.py (orchestrator)
• processors/ (document, image, audio extraction & chunking)
• vector_stores/ (ChromaDB, FAISS adapters)
• Persistence: file_storage.db (SQLite), vector_db/ or faiss_db/ directories
• docker/: Dockerfiles and compose for containerized deployment
Extract insights from README
• Processing pipeline: extract → chunk → embed → store
• Query pipeline: embed query → semantic search → retrieve contexts → LLM inference → response
• Tech stack: Python, Streamlit, SQLite, Ollama, PyTorch, Transformers, Whisper, BLIP, Tesseract, ChromaDB, FAISS