I've been building Job Hunter Team — a team of autonomous AI agents that runs a job search for you. You set the direction; they comb the job boards around the clock, read each posting, score how well it fits your profile (0–100), and draft a tailored CV + cover letter for the ones worth applying to. Fewer applications, but targeted — the final "send" is always your call.
It's MIT open source, and I'm looking for feedback, contributors, and beta testers.
👉 Repo: https://github.com/leopu00/job-hunter-team
Why I built it
I was job-hunting in early 2026 and most applications got no reply. I wired a few LLM agents together to do the tedious half of the search; in two weeks it analyzed ~200 openings, prepared ~20 tailored applications, and got me 5 interviews (caveat: my field is in high demand, so treat that as an anecdote, not a benchmark). It worked well enough that I rebuilt it properly for anyone.
It's deliberately not a mass-apply bot
The market is already an arms race — too many generic applications, so employers filter with AI, so everyone gets less attention. Job Hunter Team bets the opposite way: find the right match, and help you adapt what you offer to what the market wants. It puts AI on the applicant's side, instead of being one more filter that screens people out.
The interesting engineering: cost
An always-on team of agents running around the clock on pay-per-use APIs would be absurdly expensive. So a Bridge + Sentinel layer monitors usage live and paces the team to run for a whole month within a flat budget, instead of burning through it in hours.
One real month-long run: 658 positions found, 307 scoring 70+ (avg 71/100), across 24 countries, with no human steering. The numbers and charts are in the repo.
Architecture
Always-on core: Captain (coordinator), Sentinel (budget watcher), Assistant (chat), Mentor (career coach).
A pipeline the Captain scales on demand: Scout → Analyst → Scorer → Writer → Critic.
Each agent runs on one of three agent CLIs — Claude Code, Codex, or Kimi — with tmux + SQLite for shared state.
Stack
Node.js + TypeScript (CLI + orchestration)
Python (budget monitoring + provider glue)
Next.js + Supabase (web dashboard)
Electron (desktop app)
Everything in a single Docker container, so your machine stays clean.
Where it's going (and where I'd love help)
It's still early and, honestly, CLI-first for now — a desktop app for non-technical users is the biggest open piece. The thing I most want to crack: running it on fully local models so it costs only electricity — finding work shouldn't be gated by who can afford AI.
good first issue and help wanted are tagged in the repo. Feedback, contributors, and beta testers are all very welcome.
Top comments (0)