Back to Projects
Live Demo GitHub

AI Document Q&A

DocChat RAG

A production-shaped retrieval-augmented generation app that lets users upload PDFs, ask natural-language questions, and receive answers grounded in the uploaded documents with source/page citations.

DocChat RAG answer screen showing a cited response

Problem

Long PDF documents are difficult to search and even harder to trust when answers come from a general-purpose model. DocChat solves this by grounding every answer in retrieved document context and returning citations users can inspect.

Implementation Highlights

  • Built a FastAPI backend for upload, ingestion, retrieval, chat, streaming, health, and source endpoints.
  • Implemented PDF chunking and embedding into a persistent Chroma vector store.
  • Added a provider abstraction for OpenAI and Google Gemini LLM/embedding providers.
  • Used a relevance guardrail so out-of-scope questions are refused instead of hallucinated.
  • Returned inline citations with source filename, page number, and snippets.
  • Included an evaluation harness for retrieval quality, answer correctness, and faithfulness.

Result

The app demonstrates full-stack AI engineering: a Next.js/React interface, a Python AI backend, streaming UX, document-grounded answers, audit-friendly citations, and deployment as a single container for a custom subdomain.