← Back to work

Job-Intelligence Platform

Personal · 2025 — Present

A multi-source job aggregator with a "truth gate" for LLM résumé and cover tailoring that blocks unsupported claims — grounded in an evidence bank.

15,158
lines of code measured
445
automated tests measured
28,701
postings ingested measured
  • Python
  • FastAPI
  • SQLite
  • pytest
  • async ETL
  • LLM tailoring
  • Offline PDF export

Context

Tailoring an application to a role is useful; letting an LLM embellish is dangerous. I built a platform that does the former while structurally preventing the latter.

What I built

  • A “truth gate” for LLM résumé and cover-letter generation: every claim must ground to an evidence bank, and unsupported claims are blocked rather than smoothed over.
  • Async, multi-adapter ETL ingesting postings from many sources, with a scoring and ranking pipeline that weights fit.
  • Offline-first output — markdown-to-PDF export with no cloud dependency.
  • A real test suite — 445 automated tests over ~15k lines — because a tool that makes claims about you had better be correct.

Why it matters

It is the same idea as the RAG work, pointed at myself: grounded generation, verified before it ships.