A.I. Wrote My CV for a Real Job — Here Is Exactly How

We took a real fresher job posting (ReactJS Developer at a Bangalore startup), gave an A.I. agent a sample profile, and recorded what happened. It read the posting, scored the fit, drafted a tailored CV and cover letter, had a second A.I. agent review the drafts, revised them, compiled a clean exactly-2-page PDF, and even checked the PDF the way company screening software (ATS) reads it.

The tool is ai-job-search — open source (MIT), 15,000+ GitHub stars in four months, built on Claude Code. This is the quick start.

What you need first

• Claude Code — the framework runs inside it (a Claude subscription or API key is required; the repo itself is free).
• LaTeX for the PDF compile — lualatex (CV) + xelatex (cover letter). Minimal install on macOS:

curl -fsSL https://yihui.org/tinytex/install-bin-unix.sh | sh
export PATH="$HOME/Library/TinyTeX/bin/universal-darwin:$PATH"
tlmgr install moderncv fontawesome5 fontawesome6 academicons import luatexbase pgf \
  titlesec textpos xltxtra xunicode cite realscripts

• Optional: brew install poppler (enables the ATS pdftotext check) and bun (only needed for the job-board scraper).

The 5 steps

1. Fork and clone:

gh repo fork MadsLorentzen/ai-job-search --clone && cd ai-job-search

2. Fill your profile. Run claude, then /setup — it offers three paths: point it at a folder of your documents, paste one existing CV, or answer an interview. Everything lands in .claude/skills/job-application-assistant/01-candidate-profile.md. Honest data only — the system’s own rule is that it never fabricates skills or experience.

3. Point it at a job:

/apply <job posting URL>

If the portal blocks automated fetching (many Indian portals do), paste the posting text directly — that is documented and works exactly the same:

/apply Reactjs Developer (Fresher) — <company>, Bangalore. <full posting text…>

4. Watch the workflow. It evaluates your fit against the posting (and tells you honestly when a role is a poor deal), drafts the LaTeX CV + cover letter, spawns a fresh-context reviewer agent to critique them, revises, compiles, and iterates until the CV is exactly 2 pages. In our recorded run this took about 14 minutes end to end.

5. Review and send — yourself. The output is cv/main_<company>.pdf and cover_letters/cover_<company>_<role>.pdf. It drafts; it never submits anything anywhere. You stay in control of every application.

The India adaptation (read this)

The repo ships job-board scrapers for Danish portals — those are not useful here. What works for India:

• linkedin-search is country-agnostic: linkedin-search -l "Mumbai, Maharashtra, India" (personal use only, per LinkedIn’s terms).
• /add-portal scaffolds a scraper skill for a local board — point it at Naukri or any portal you use and it generates the integration for you.
• Or skip scraping entirely: find postings yourself and use /apply with the URL or pasted text, like we did.

Useful extras

/rank (compare multiple postings), /interview (prep from your profile + the posting), /outcome (track application results), /upskill (gap analysis). Note: this is a community project, not an official Anthropic product.

The honest summary

A tailored CV is not magic — it is work that most people skip because it takes hours per application. This makes it take minutes, with a reviewer pass and an ATS check built in. Recruiters can tell a copy-paste CV. A tailored one gets read.