# Document Copilot

An example internal AI chatbot, built for a fictional investment research firm, that lets analysts query a corpus of financial filings in plain English and get sourced answers.

Document Copilot is a demonstration project for an internal AI chatbot that lets analysts query a corpus of documents in plain English and receive sourced, citable answers, built around a fictional client, a fictional independent investment research firm whose analysts spend much of their week reading regulatory filings before producing original analysis.

Its stack combines a Python and FastAPI backend, a Vite, React, and TypeScript frontend, a Supabase Postgres database for users, chats, documents, and text chunks with SQLAlchemy and Alembic migrations, retrieval through Supabase's pgvector extension plus Postgres full-text search, Supabase email authentication, and OpenAI for both the language model and embeddings, hosted on Railway.

It includes a standalone downloader that fetches a small sample of real SEC EDGAR 10-K filings for a handful of large public companies to use as sample data, and ships with a written client brief describing the fictional scenario the project is built around.

Website: https://github.com/daveebbelaar/document-copilot