The problem
Job hunting is slow and annoying. I’m not looking for any job. I want a company and a role where we’re both better off for it, and that’s hard to find without help. Job boards sort by keyword and date. I wanted something that sorts by how well a job actually fits me.
What it does
Every morning Job Radar:
- Gathers about 16,000 postings from 8 sources: company job boards and public job APIs.
- Scores each one against my profile.
- Publishes the best ~1,000 to a private site I can use from my phone like an app.
From there I can save, hide, or track applications, and the tracker syncs between my phone and computer. For a job I like, “Help me apply” has Claude draft a short cover letter that I edit and send myself.
How the scoring works
I wanted scoring I could explain, not a black box:
- Pay and resume fit count most. It reads my uploaded resume, matches it against about 150 skills, and shows “Matches your resume” and “Also asks for” on every job.
- Title matters less. Role types sit in three groups (Top, Core, and Adjacent) so applied AI and data science roles rise to the top.
- Real-life filters. Remote and hybrid jobs get a boost and fully onsite jobs a penalty. Jobs asking for far more experience than I have are marked “Stretch” or pushed down.
- AI-coding signal. Jobs that mention building with AI tools get a small bump, because that’s where I’m headed.
How I built it
I built it with Claude Code, working as the product owner. I set the rules up front, and every new feature started as a written plan I approved before any code was written:
- Only sources that allow it. No scraping of sites that forbid automated access.
- Free unless it’s really worth paying for. The only paid piece is the cover letter feature, with a daily cap and spend tracking in the code.
- Nothing personal on the site. My resume never goes in the code, contact details are stripped on upload, the site sits behind a login, and tests check every published page for personal data.
- Verify, don’t invent. Every API limit and term of service is checked against the provider’s docs.
It came together much faster than Resonance. A month earlier I had never used Git or written a test. By this project, planning, testing, and committing were habits.
Fixing the AI when it fell short
The first cover letters Claude wrote were correct, but they sounded robotic, the kind of letter a hiring manager spots as AI in one line. So I treated it like a product bug. I gave the model a formula for how a good letter is built and added “My stories”, a short set of real experiences in my own words for it to draw from. The letters now start from something true about me instead of generic praise for the company.
Results
- A daily, ranked list I actually trust, in place of hours of scrolling job boards.
- Applying is exciting now. The research is thorough enough that every job on the list is one I can picture myself at.
- A real, working AI product: a pipeline, a scoring model, a synced app, and a Claude-powered feature with cost controls, all built and used by me every day.
What I learned
- Don’t lose sight of the goal. I built this because I’m looking for a job. With AI, I could have spent every day adding features and making it cooler, and that would have felt productive. But the goal was never to build a job radar. It was to apply for jobs. When AI makes building this easy, the most important thing is to set the goal clearly at the start and keep checking against it.
- Set the rules before the code. Writing down privacy, cost, and source rules first meant Claude could move fast without me worrying about what it might do.