A geophysicist working in Denmark was laid off at the end of 2025 and built himself a job-hunting pipeline in Claude Code. He sent 69 tailored applications, landed 20 first-round interviews, signed one contract, and in June 2026 switched careers to become an AI engineer. He open-sourced the whole pipeline as AI Job Search, and over the past few days it has been gaining more than 2,000 GitHub stars a day on the trending page, passing 53,000 in total.
I went through the README, SETUP guide and the release notes for the last two versions. The short version: this isn’t a job app you open and click through. It’s a Claude Code workflow that runs on your own computer. If you’re comfortable with the command line and already pay for Claude, it’s a very smooth way to tailor résumés, write cover letters and prep for interviews. If what you want is one-click mass applying on a job board, it won’t help. Details checked on October 19, 2026.
What it actually does
Once installed, you run a job search through a handful of slash commands inside Claude Code:
| Command | What it does |
|---|---|
/setup |
Builds your profile: reads the résumé PDFs, LinkedIn export, certificates and reference letters you drop into documents/, or a pasted résumé, or interviews you question by question |
/scrape |
Searches job sites for roles that match your profile, removes duplicates and ranks them by fit |
/rank |
When the search returns too many jobs, scores them in bulk and hands you a ranked shortlist |
/apply <link or job description> |
Assesses fit, rewrites your résumé and cover letter for that role, sends a second agent to research the company and critique the draft, then outputs PDFs |
/interview |
Builds a prep pack for a specific round: the original job ad, the résumé you sent, likely questions, plus mock interviews |
/outcome |
Logs the result of each application; for anything with no reply after 10 days, drafts a follow-up email (draft only, it never sends for you) |
There’s also /upskill (compares your skills with a target role and suggests a learning plan), /html-report (an offline application dashboard), /gmail-sync (spots interview invites and rejections in Gmail and lists them for you to confirm), /notion-sync and more.
How it differs from the usual “AI résumé rewriter”
Most AI résumé tools take a pasted job description and spit out one version of your résumé. AI Job Search does a few genuinely useful things on top:
- Writing and reviewing are separate. One agent writes the first draft; another, with a fresh context, researches the company and picks holes in it; then the writer revises. That catches a lot of the missing keywords and filler that single-pass generation lets through.
- It actually looks at the PDF. The résumé is typeset in LaTeX, and after compiling, Claude inspects the rendered pages and keeps adjusting until the résumé is exactly 2 pages and the cover letter exactly 1, with no heading stranded at the bottom of a page. v1.7.2 added a script that measures white space and page breaks.
- It checks from an ATS’s point of view. Many companies screen résumés with an applicant tracking system (ATS), which reads the PDF’s text layer, not the layout you see.
/applyextracts that text layer, checks that your contact details are plain text and the reading order makes sense, then compares it against the job’s keywords. - It doesn’t invent experience. The README is explicit: every line in the résumé and cover letter must trace back to something in your profile, and skills the job asks for that you don’t have are flagged as gaps rather than slipped in.
That last point matters most to me. The scariest thing about letting AI write your résumé is it “inventing” a project you never did.
Can you use it outside Denmark?
Yes, with a few rough edges to know about:
- The built-in job sites are Danish. The bundled search plugins cover four Danish boards (Jobindex, Jobnet, Akademikernes Jobbank and Jobdanmark) plus two that work anywhere: LinkedIn’s public job listings (the author warns that automated access breaks LinkedIn’s terms, so keep it to light personal use) and the tech-job aggregator freehire.me.
/add-portalcan generate a plugin for your local job site, but it refuses sites that require a login. So for most people, the practical route is to skip/scrapeand paste the full job description into/apply, which the README supports. - The templates assume an English-style CV. The default résumé uses moderncv (LaTeX) and the cover letter uses the bundled cover.cls. For a different look, or a résumé in another language, swap in your own LaTeX or Typst template with
/add-template. I didn’t test non-Latin scripts, so compile a sample first. - It needs Claude Code, and Claude Code isn’t free. Claude Code comes with paid Claude plans: Pro is $20 a month billed monthly (about $17 a month billed yearly), or you can pay per use through the API, which the author says is usually cheaper if you only use it occasionally. If you use another agent like Codex or Gemini CLI, start from the repo’s AGENTS.md: the job-site search plugins work as-is, but for the full workflow you’ll need a community-adapted version.
Getting started
Install these first: Claude Code, Python 3.10 or later, Bun, and a LaTeX distribution with lualatex and xelatex (TeX Live, MacTeX, TinyTeX or MiKTeX all work). For the ATS check, also install pypdf.
# 1. Get the code (see the privacy note below; a private repo is recommended)
git clone https://github.com/MadsLorentzen/ai-job-search.git
cd ai-job-search
# 2. Install the job-site search tools (on Windows: python tools\install_portal_clis.py)
python3 tools/install_portal_clis.py
# 3. Start Claude Code and build your profile
claude
/setup
# 4. Generate a résumé and cover letter for one job
/apply <paste the full job description here>
The more detail in your profile, the more the output sounds like you. The author’s advice: don’t just list job titles. Write what you actually did in each role, which tools you used and any results you can put numbers on.
Two easy mistakes to avoid
- Don’t just fork it. Forks of a public GitHub repo have to stay public, and
/setupwrites your name, contact details, work history and salary expectations into files that get committed. If this is for your own job search, follow section 8 of SETUP.md to create a private repository and set the original as upstream. - Job pages can hide “instructions” too. The author treats scraped job ads as untrusted content: the workflow won’t follow instructions inside them or open links in the text. But he also says this is instruction-level protection, not a sandbox. On unfamiliar job sites, read what it scraped and what it wrote before you send anything.
Who it’s for
- Good fit: people already using Claude Code who are about to apply to a lot of jobs, especially roles that need an English résumé and cover letter; anyone who wants a record of every application and interview round to learn from.
- Not a fit: people who don’t want to touch the command line; people who mostly apply through one-click buttons on job platforms; and anyone sending two or three applications a month, who doesn’t need such a heavy process.
New to the tools around Claude Code? Our overview of the AI coding agent ecosystem (in Chinese) explains what skills, MCP and the rest each do. For more Claude Code add-ons in English, see Coucou and Magpie.
Project info
- Latest version: v1.7.2 (released September 30, 2026, Beijing time); the repo is still updated frequently
- License: MIT, free and open source
- Runs on: Claude Code on Windows, macOS or Linux (needs a paid subscription or API credits)
- Dependencies: Python 3.10+, Bun, LaTeX (lualatex + xelatex), optional pypdf
- Stars: about 53,000 (October 19, 2026)
- Repo: MadsLorentzen/ai-job-search
- Full setup: SETUP.md
- Claude pricing: official Claude pricing page
