Built with AI
Job Hunt Automation
My own job search, turned into a tool. It finds roles, judges them against what I'm actually looking for, scores my CV against each posting, and drops only the survivors into a private dashboard for me to review.
4
Parts working together
2+
Models judging fit
CV
Scored against each role
1
Dashboard to review it all
The problem
A job search is mostly repetitive judgment. You find roles, bin the obvious mismatches, weigh what's left against what you actually want, then try to remember weeks later why you passed on something.
None of that is hard once. It gets painful at volume, when the same marginal fits keep circling back and you keep spending slow, expensive model time re-deciding things you already decided.
What it does
- Pulls roles from a few different sources and tidies them into one shape.
- Skips anything it has already seen, rejected, or that's already in the dashboard, before any model runs.
- Judges fit with a mix of local and cloud models, depending on the workflow.
- Scores my CV against each posting with an ATS-style scanner I built with AI.
- Sends the survivors to a private dashboard where I move each one through application states.
How I built it
None of this needed an engineering team. I decided how it should behave and let AI write most of the code, the ATS scanner included. The judgment calls were mine: mostly what to skip, and what counts as a good-enough match.
The piece I'm proudest of is the reject ledger. Once I've turned a role down, it never gets judged again, which saves a real amount of GPU time and cloud spend, plus a lot of my patience. I also kept the dashboard separate from the scraping and judging code, so I can swap the model behind the judging without disturbing how applications get tracked.
Built for me, not for sharing
Some of what I build is meant to go public, like Metro Agora. This one isn't. It runs on my own career memory: job reports, rejected roles, CV-fit assumptions, the dashboard's data. That's the entire point of it, and the reason it stays private. What's worth showing is how it's built. The contents stay mine.