AI Job Search is a job application framework that runs inside Claude Code, turning the coding agent in your terminal into an assistant for your own job hunt. It is aimed at people who are comfortable forking a repository, editing text files and living in a command line, and who are applying often enough that rewriting the same cover letter by hand has become the bottleneck. You fork the repo, fill in your profile once, and then hand it job postings one at a time.
What it does
Point it at a job posting and it works through the application:
- scores how well the posting fits the profile you filled in
- drafts a CV tailored to that specific posting, in LaTeX
- writes a matching cover letter
- runs a reviewer agent that critiques the draft and revises it before it ships
- prepares you for the interview
The README describes the loop as three commands used weekly: /scrape to bring in postings, /apply to produce the documents, and /interview to prepare. Your profile and your generated documents live in your own fork, on your own machine, rather than in somebody's web app.
How it works
The repository is not a model and not an application. It is a set of commands, prompts and document templates for Claude Code, the agent Anthropic ships for the terminal. Claude does the reading and the writing; the repo decides what it gets asked and in what order. Python is the main language, and the documents are LaTeX, which is why the output is a typeset CV rather than a text box.
The piece worth noticing is the review step. The first draft is not what you send: a second agent pass criticizes it and then rewrites it. That is the structural difference between this and one long prompt in a chat window.
The author is explicit that the project is independent and not affiliated with, endorsed by or sponsored by Anthropic, and that it has no token, coin or paid sponsorship program. Anything claiming otherwise, he writes, should be treated as a scam; the only ways to support it are a Ko-fi link and contributing on GitHub.
Getting started
The code is free, under the MIT license. The intelligence is not. The framework runs on Claude Code, so you need either a Claude subscription or paid API credits, and that is a real running cost on every application you generate. The materials name no local-model option, so there is no way here to avoid paying for the model.
Installing is a fork, not a package manager command: fork github.com/MadsLorentzen/ai-job-search, fill in your profile, then drive it from Claude Code with the repo's commands. There are no release builds to download and no installer. Because the documents are LaTeX, you will also want a LaTeX toolchain on hand to turn the drafts into PDFs. Beyond that, the project doesn't list hardware requirements: no GPU, no minimum memory, no supported operating systems, which is what you would expect from something that is mostly prompts, templates and text files.
When to use it / when not
The main catch is cost and shape. You pay a model provider for every application you produce, and you are maintaining a fork of someone else's prompt framework, in a terminal, with LaTeX in the pipeline. None of this is a polished product you log into.
Use it instead of pasting the posting and your CV into a Claude or ChatGPT window when you are applying often enough that the repeatable parts start to matter: one profile of record, consistent templates, a fit score before you spend an evening on a long shot, and a critique pass you do not have to remember to ask for. Don't use it if you send a handful of applications a year, or if editing config files and LaTeX sounds worse to you than simply writing the letter.
About the headline numbers: sixty-nine tailored applications, twenty first interviews, one signed contract, and a start as an AI engineer in June 2026. That is the author's own account of his own search, after his geophysics position was cut in late 2025, and it is the only evidence offered. It is not an independent study, and there is nothing for a reader to reproduce. He adds that he told every employer he was using the framework, and that it usually started a technical conversation rather than counting against him.
Take this seriously if you are technical, job hunting or about to be, and already paying for Claude. In that situation the framework is close to free to try and gives you something a chat session cannot: a stored profile, versioned documents, and a fixed sequence that always includes a critical second look. Nearly 48,000 stars and more than 16,500 forks suggest the format is working the way it was meant to, since every serious user ends up with a private copy that holds their own career file. If you want a push-button job search with no setup and no bill, this is not that. If you want to own the machinery and can read Python and prompts, it is a strong skeleton to start from.