Flagship Project
Enterprise RAG Platform
Production-style enterprise RAG platform for document ingestion, hybrid search, grounded chat, streaming responses, and citation-backed answers.
A full-stack retrieval-augmented generation system that ingests enterprise documents, extracts and chunks text, creates embeddings, stores vectors in PostgreSQL with pgvector, and answers questions using locally hosted Ollama models. The platform includes a FastAPI backend, Streamlit UI, persisted chat sessions, hybrid vector and keyword retrieval, streaming chat, and auditable source citations.
Tech Stack
- Document upload and ingestion
- PDF, text, and markdown extraction
- Paragraph-aware chunking with overlap
- Hybrid vector and keyword search
- Persisted chat sessions
- Streaming chat UX
- Citation-backed answers
- Document and chunk management
- Docker Compose stack with FastAPI, Streamlit, PostgreSQL, pgvector, and Ollama
- Layered backend with API routes, services, repositories, schemas, and database models
- PostgreSQL stores documents, chunks, embeddings, chat sessions, and message history
- pgvector IVFFlat and Postgres full-text indexes power hybrid retrieval
- RAG workflow returns grounded answers with source citations