Document Intelligence API

Overview

An asynchronous REST API that parses, chunks, and vectorizes documents (PDF, DOCX, CSV) for LLM-powered Q&A. Built with FastAPI, Celery, and pgvector. A document intelligence system that allows users to upload files to a MinIO object storage bucket. The system uses a Redis-backed Celery worker to asynchronously parse the documents, split them into optimal chunks, and generate 3072-dimensional vector embeddings via the Gemini API. These embeddings are indexed in PostgreSQL using pgvector. The system features a semantic search endpoint that retrieves relevant document excerpts to ground LLM-generated answers with accurate citations.