Document Intelligence Pipeline

Retrieval and extraction across twenty years of unstructured engineering reports, with citations back to the source page.

Client
Placeholder client
My role
Technical Lead
Year
2026
Category
AI Solutions
  • Python
  • LangChain
  • pgvector
  • Redis
  • AWS

Two decades of PDFs, scanned in varying quality, with no consistent structure and no metadata worth the name.

What mattered

Chunking strategy, far more than the choice of embedding model. Reports written as continuous prose needed different handling from tabulated test results, and treating them identically produced answers that were fluent and wrong.

Citations first

Every answer links to the page it came from. Engineers do not trust a system they cannot check, and they are right not to.

Outcome

Searches that used to take an afternoon now return a cited answer in seconds.

Related work

Clinical Triage Assistant

A triage support tool that ranks incoming cases by urgency and, crucially, abstains when it is not confident enough to be useful.

Read more

Industrial Inspection Vision System

Surface defect detection running on the line itself, trained from a few hundred labelled examples rather than a few hundred thousand.

Read more