DocIntel

Document Intelligence Hub  ·  Local RAG  ·  Full-Stack AI

DocIntel Document Q&A
DocIntel Entity Extraction

Project Info

  • Domain: GenAI / RAG / Full-Stack AI
  • Stack: Python, Ollama (Llama 3.1 8B), ChromaDB, BM25, FastAPI, React, Vite, spaCy, PostgreSQL
  • GitHub: View Repository

The Problem

Businesses that work with sensitive documents — contracts, legal filings, financial reports — cannot send them to cloud LLM APIs. Existing document AI tools force a trade-off between intelligence and data privacy.

What I Built

  • Built a fully local document intelligence system (zero external API calls) supporting Q&A with page-level citations, structured summaries, NER, and side-by-side document comparison — all powered by Llama 3.1 8B via Ollama
  • Implemented hybrid retrieval combining dense vector search (ChromaDB + nomic-embed-text) and BM25 sparse retrieval with Reciprocal Rank Fusion, improving recall on complex business queries
  • Full-stack React + Vite frontend backed by FastAPI and PostgreSQL; entity extraction surfaces people, organizations, dates, monetary values, and contract clauses as structured tagged output

Impact

Enables teams to deploy powerful document AI on-premises — no data leaves the machine, making it viable for regulated industries where cloud LLM use is restricted.