Skip to content
← all work
✻ case study · run #01

Document Intelligence Platform

Genworx · SDE-1Python, LangGraph, PostgreSQL/pgvector, Gemini & Claude on Vertex AI, Azure AI Document Intelligence

run #01 · genworx✻ AI

Document Intelligence Platform

A headless platform that reads contracts, invoices and almost any business document, and returns structured data.

pipeline
contract.pdf
invoice.pdf
scan.png
summarytablesimages
{
  "type": "invoice",
  "vendor": "Acme Ltd",
  "total": 1240.00,
  "lines": 6
}
10docs per batch, in parallel
3specialist agents
2LLM providers

Problem

Businesses receive contracts, invoices and many other document types. They need the data inside them, not the PDFs.

What I built

  • Parsing with Azure AI Document Intelligence; extraction with Gemini and Claude on Vertex AI.
  • Summarization, table-processing and image-processing agents orchestrated with LangGraph.
  • Retrieval-augmented inference over PostgreSQL/pgvector.
  • Parallel, worker-based batch processing: 10 documents per batch, concurrently, with failed steps retried on their own.

Architecture

Upload → Parse → Agent graph (summary / tables / images) → RAG over pgvector → Structured output