./job-application-agent/

Built a proactive job-hunting agent that scrapes or ingests job postings, parses company/title/requirements, scores fit against a structured profile and project bank, and produces versioned resume and cover-letter artifacts on demand.

The backend is a FastAPI service with token auth, CSV tracker persistence, and pipeline orchestration shared with a Typer CLI. A static React/Vite dashboard reviews discovered jobs, filters by fit/company/search, stars rows, bulk archives low-signal postings, and opens per-job resume, cover-letter, and audit views through the same API.

Designed the system around truth constraints: resume bullets are selected from approved YAML facts, PDF rendering re-runs the claim auditor before output, and cover letters pass a separate allowlist audit for unsupported technical claims. Failed cover-letter audits still preserve the draft, metadata, and audit JSON on the dashboard, while PDF rendering remains blocked until the claim check passes.

Each tailoring run writes monotonic per-job artifacts - markdown, HTML, PDF, metadata, and audit reports - so earlier resumes and cover letters are never overwritten. The tracker also records application outcomes, response dates, interview stages, source quality, and backup/restore snapshots so the system can measure which sources actually produce responses instead of only counting applications.

A scheduled scraper can run on a small GCP VM, pulling Greenhouse, Lever, Ashby, Workday, and Hacker News postings every four hours through a Cloudflare Tunnel into the local backend. Discovery is idempotent, high-fit roles can trigger Discord notifications, and the dashboard keeps the human review step in front of every resume, cover letter, and application status change.