AI Job Search

• 1. overview (709 chars)

  AI Job Search is an open source AI job search framework built on Claude Code and written by Danish developer Mads Lorentzen for himself after he lost his job at the end of 2025. The framework breaks down the job search process into programmable workflows—from profile setting, job search matching, to dual-agent review generation of resumes and cover letters, PDF formatting verification, to interview preparation and results tracking—all driven by Claude Code’s slash commands. After the project was released at the end of March 2026, it quickly shot to the top of GitHub Trending. As of the end of July, it had received more than 19,000 stars, making it one of the fastest growing Claude Code workflow projects. It does not click "submit" for you, but completes all the repetitive work from evaluation to finalization of each application for you.

• 2. Product background (1662 chars)

  Mads Lorentzen is a Ph.D. in geophysics. After being laid off by his company at the end of 2025, he fell into a cycle of repeatedly revising his resume and writing cover letters. He found that repeating the same process for every job application—reading JDs, assessing fit, tailoring resume wording, writing cover letters—was time-consuming and easy to rush through. For a developer who is familiar with Claude Code, he spent three months writing his ideal job search process into a reusable framework, named ai-job-search. The project was first submitted to GitHub on March 23, 2026, and began as a simple prototype. What really ignited the community was the author’s “self-use success story”: he used this framework to apply for 69 applications, received 20 first interview opportunities, and finally successfully joined the company as an AI engineer in June 2026. The results went viral after being shared on LinkedIn and Reddit, with many people realizing that this wasn't just a "resume builder" but a complete job search operating system. The project is completely open source using the MIT protocol and is not associated with any commercial companies or cryptocurrency projects. The author expressly states that this is not an official product of Anthropic and no fee was charged for its use. The framework currently uses TypeScript, Python and LaTeX as the main languages ​​and is hosted under the MadsLorentzen/ai-job-search repository. The core of the entire project is not the amount of code (currently only about 55 commits), but the integrity of the workflow design and engineering thinking. Among similar competing products, career-ops prefers batch pipeline processing (multiple workers in parallel, Go TUI dashboard), while AIHawk focuses on automated form filling on web pages. AI Job Search is caught in the middle and has chosen the "quality first" route - it does not pursue delivery quantity, but pursues the ultimate customization of each application.

• 3. Product functions and usage experience (3360 chars)

  The core concept of AI Job Search is to "break down job hunting into programmable workflows." It does not provide a web interface, but drives the entire process through nine slash commands within the Claude Code command line. Each command corresponds to a key link in job hunting. The three most basic commands are /setup, /scrape and /apply. /setup is used to create a personal profile and supports three paths: if you already have a documents folder (containing resume PDF, LinkedIn exported files, academic certificates, etc.), Claude will directly read it and organize it in a structured way; if you only have a ready-made resume, you can paste it into the conversation and let Claude import it; if you have nothing, the system will gradually ask about your background, skills and career goals through a structured interview. The author particularly emphasizes that the more detailed the file, the higher the quality of subsequent output - it is not a list of skill names, but a description of what tools you used in what scenarios, what you did, and what results you achieved. /scrape is responsible for automatically searching multiple job portals, sorting them by matching degree after removing duplicates. The current built-in search skills are mainly for the Danish market (Jobindex, Jobnet, Akademikernes Jobbank, etc.), and also include a zero-runtime search tool based on LinkedIn's public endpoints. The author's design intention is very clear: this mode itself is to replace the local portal for you. The /add-portal command allows you to generate search skill plug-ins for job portals in any region, and the community can also contribute adaptations. In addition, there is a /rank command, which performs batch and parallel scoring on the captured positions, scores them from the five dimensions of skill matching, experience matching, cultural fit, geographical location, and career development consistency, and generates a sorted short list. /apply is the core command of the entire framework and the one that best reflects the engineering level. When you enter /apply plus a job link in Claude Code, the system will first parse the job description (treat the job content as an untrusted input and will not execute any instructions in it), and then perform a multi-dimensional matching assessment with your personal profile. Next, enter the most important "drafting-review" dual-agent workflow: the first Claude agent serves as the drafter, using LaTeX templates to customize resumes and cover letters based on the evaluation results; then the second Claude agent starts with a new context window, independently researches the target company from the perspective of a "picky reviewer", and criticizes the first draft - checking for missing keywords, weak expressions, overly general language, etc.and a focus that does not match the job requirements. The drafter will make revisions after receiving review comments. This cycle is the most fundamental difference between AI Job Search and other resume generation tools. After the text content is completed, the system also has a unique PDF verification loop. It uses lualatex to compile the resume, xelatex to compile the cover letter (because cover.cls relies on fontspec), and then reads each page of the PDF to check the layout - whether the resume is exactly two pages, whether the cover letter is exactly one page, whether the title line does not fall to the bottom of the next page in isolation, whether the font is consistent, whether the signature is visible. After problems are discovered, LaTeX parameters will be automatically adjusted and recompiled until the output is satisfactory. Finally, use pdftotext to extract the PDF text layer, simulate the parsing effect of the ATS system, and check whether the contact information is complete and the keyword coverage is good. This validation loop solves the most common and vexing problem with LaTeX resumes: inconsistent formatting across different compilers. Many

• 4. Pricing and business model (768 chars)

  AI Job Search itself is a completely free open source project, using the MIT license. The cost of using it comes mainly from Claude Code's subscription fee ($20 per month) and the token consumed each time the /apply command is run. Since it involves multiple interactions between two agents, the drafter and the reviewer, the cost of a single application is not low. This is a typical "token for quality" model.   The project currently has no commercialization plan, no charging levels, and no paid value-added services. The author has clarified multiple times that the project does not have any associated tokens, NFTs, or paid sponsorship programs. Its path to monetization is more indirect—the author himself demonstrated his engineering capabilities through this project and eventually found a job. For contributors, participating in such a high-profile project also brings professional exposure in itself.

• 5. User reviews (2089 chars)

  Community feedback on this project has focused on a few recurring themes. The most praised is the dual-agent review design. Users on Reddit generally believe that having a second agent review the first draft from an "adversarial" perspective is the key design difference between "usable AI output" and "fake AI copywriting at a glance." Many HR practitioners on LinkedIn also approved of this design, believing that it was at least aware of the risks of AI-generated materials and tried to mitigate them through review mechanisms. Authenticity guarantees also received positive reviews. All claims made by the system must be based on real experiences in the user profile and never fabricated skills or experience. ATS keywords also add content that is actually supported in the user profile, rather than randomly inserting popular keywords. The community believes that this is an important safety net for users who are worried that AI-assisted job applications will further deteriorate the integrity of the job market. Additionally, the PDF validation loop is widely considered to be the framework’s most unique selling point. Multiple users said that the content generated by other AI resume tools often "reads as a bunch of gibberish" in the ATS system - contact information is recognized incorrectly, and content in multi-column layouts is lost - while AI Job Search has at least made systematic checks and fixes in this regard. However, LaTeX dependency has been repeatedly mentioned as the biggest barrier to use. Job seekers with non-technical backgrounds—precisely the people who need the most job search help—often give up when faced with TeX Live installation, fontspec compatibility, and the differences between lualatex and xelatex. There are several community forks trying to replace LaTeX with simpler HTML/PDF paths. In addition, the Danish market-first design also means that non-Nordic users need to use /add-portal to adapt themselves, or fall back to the process of pasting job descriptions, which will lose the convenience of automatic crawling. Some users have also expressed concerns about the long-term impact of AI’s mass customization of resumes. "If job seekers use AI to write deeply customized resumes, and recruiters also rely on AI to screen these resumes, then what is the point of the entire resume process?" - Although this question is not specific to AI Job Search, it does reflect that AI is reshaping the game between job seekers and recruiters.

• 6. Industry evaluation (1303 chars)

  The developer community's evaluation of ai-job-search focuses not only on the product itself, but also on the model it represents. AgentConn's analysis article pointed out that the dual-agent review design is an architectural pattern that can be generalized - any high-risk agent workflow (contract drafting, grant application, customer presentation) can benefit from the design of "the second agent conducts adversarial review in an independent context".   There are many in-depth analysis articles on Nuggets and CSDN, which systematically compare the project with career-ops and AIHawk. The conclusion is that the three have different focuses - AI Job Search takes the "quality first" route and does the most thorough work in the two aspects of typesetting inspection and ATS verification. The AI ​​tool set ai-bot.cn lists it as a representative of the "open source intelligent job search framework" and believes that its design reflects the evolution trend of AI applications from "question and answer tools" to "workflow agents".   Professional discussions on LinkedIn are also very popular. Many practitioners in the field of recruitment believe that this project reveals a deeper trend: job hunting is changing from a "writing problem" to a "process problem" - AI handles the first repetitive work, and humans use their energy where judgment is really needed. Others point out that the framework itself may be more persuasive than any resume it generates—the fact that the author was hired on the basis of this project is a prime example.

• 7. Controversies and Risks (976 chars)

  The core risk comes from third-party platform restrictions. Automatically scraping LinkedIn job information violates LinkedIn's terms of service, and the project team clearly recommends in the README to keep the frequency low and for personal use only. The anti-crawling strategies of different job portals are also constantly evolving. These external vulnerable links are not problems that can be solved by code.   In terms of security, the design of the project takes into account the risk of prompt injection - treating job descriptions as untrusted input and not executing the instructions therein. But the author also admits that "the agent-level defense is at the command level, not at the sandbox level." This means that when using it on an unfamiliar recruitment portal, it is recommended that users check the crawled content and generated application materials before sending.   Additionally, generating highly customized resumes at scale can trigger counterattacks from ATS systems. Recruitment platforms have begun to optimize filtering algorithms for AI-generated content. Over-optimizing resume keywords may trigger anti-fraud flags.

• 8. Summarize (596 chars)

  AI Job Search is one of the most impressive AI workflow projects in the first half of 2026. Its value does not lie in the advanced technology it uses, but in its deep enough understanding of "job hunting" - it is not a tool that "helps you write a resume", but it reduces job hunting to a complete closed loop including evaluation, customization, review, verification, tracking, and feedback. The two designs of dual-agent review and PDF verification loop make it a unique position among many AI resume tools. For Claude Code users, this is an almost must-try workflow. For those looking for a job, especially those with a technical background, it has the potential to fundamentally change the way you search for a job.

No in-depth report yet

• 

  谭素

  —

  Tried ai-job-search today. Imported my resume with /setup then ran /apply to see the result. The PDF did compile, but setting up the LaTeX environment took me nearly an hour. Non-technical users will probably give up before even getting started.

---

• 

  Dorothy_TorresIII89

  —

  The creator's story of landing a job through this framework is incredible. 69 applications, 20 interviews, 1 offer. That hit rate proves quality matters more than quantity. Already forked it, going to try it out this weekend.

---

• 

  姜明

  —

  The drafter-reviewer dual agent design is the standout feature. Having a separate agent critique your own resume feels like having someone watch your back. Way more reliable than those one-shot generation tools.

---

• 

  afy1l

  —

  Rating: functionality 9/10, setup difficulty 6/10. Is it powerful? Absolutely. But asking me to install Bun, then LaTeX, then configure lualatex and xelatex, a lot of people will drop off right there.

---

• 

  MJohnson_Plus

  —

  The ATS validation step is really useful. I used another tool to generate my resume before, and the recruiter said my contact info didn't even get parsed. AI Job Search at least verifies the PDF text layer — that's well done.

---

• 

  MargaretAnderson_2020

  —

  Works great locally in Denmark. Built-in Jobindex and Jobnet search work out of the box. But outside the Nordics you have to configure /add-portal yourself. Basically zero support for the Chinese market. Hope someone in the community makes an adapter.

---

• 

  Samuel.RuizJr

  —

  After two weeks, my biggest takeaway is that /apply doesn't just fill a template. It genuinely adjusts emphasis based on each job description. But the token consumption is no joke. Running one application feels like Claude is burning through budget.

---

• 

  WilliamKim168

  —

  Honestly a bit worried this kind of tool will distort the job market. If everyone uses AI to write deeply tailored resumes and employers use AI to filter them, what's even the point of a resume anymore?

---

• 

  ZThomas7

  —

  The security design is thoughtful. Job descriptions are treated as untrusted input and instructions within them are never executed. The README honestly admits agent-level defenses are not a sandbox though, so be careful on unknown portals. That transparency is actually reassuring.

---

• 

  Shirley.Davis_88

  —

  I compared it with career-ops and AIHawk. career-ops leans toward batch pipeline style — 200 applications in three days. AI Job Search is a precision approach. Depends on what you want. I'd pick the latter.

---

• 

  INguyen_2023

  —

  The PDF validation loop is genius. When I write LaTeX resumes manually, my biggest fear is the layout breaking after compilation. It automatically checks page count, orphaned headings, and font fallbacks — solving the biggest pain point for LaTeX users.

---

• 

  MsRichardWest_x

  —

  The 19K-star growth is no joke. This project represents a trend — AI applications evolving from 'Q&A tools' to 'workflow agents.' It doesn't just chat a bit more; it actually gets things done.

---

• 

  Jason282

  —

  First time running /scrape after setup. Six portals searched in parallel with dedup and ranking. Results were clearly presented. But the LinkedIn search uses public endpoints, which is a gray area. Fine for low frequency personal use, definitely risky at scale.

---

• 

  cHAINbASE

  —

  What impressed me most is the principle of 'never fabricate skills.' Many AI resume tools stuff your resume with things you haven't done, but this one uses a reviewer agent to verify authenticity — it holds the line.

---

• 

  trueCaitlinChen_pro

  —

  To be honest, this project only has 55 commits. The value isn't in how many lines of code were written, but in the depth of understanding about the job-seeking process. Breaking the whole workflow into nine commands, each corresponding to a decision point — the design is incredibly clear.

---

• 

  Lucas58

  —

  I tried it after seeing Eduardo's post on LinkedIn. 53 comments with HR folks and developers arguing. Some call it a revolution in job search efficiency. Others say it just adds noise for recruiters. I think the tool itself is fine. It is all about how you use it.

---

• 

  梅花_9

  —

  Just ran /apply, and the generated resume layout looks way better than what I hand-crafted. The moderncv banking style with LaTeX's polish is a real plus when applying to the European market.

---

• 

  SButler_2024

  —

  The creator is a PhD geophysicist who coded his way to a job after being laid off. The story itself is compelling. Him getting an offer through this framework isn't the point. What matters is he demonstrated the commercial value of agent workflows through real results.

---

• 

  星辰_17

  —

  Finally someone automated the pain points of LaTeX resumes. Before, every time I edited a .tex file and compiled to PDF, I had to manually check headers and footers. Ai-job-search does this check for me directly. Saves a ton of time.

---

• 

  Emily_Henderson_66

  —

  The framework is most friendly to Mac users. Linux is okay. Windows users have to mess with WSL. I think the creator should make a Docker image for one-click deployment. The current barrier really blocks most people.

---

• 

  J_ulian276

  —

  18K stars with 3,700 in a single day, hitting number one on GitHub Trending. This popularity shows how much developers need this kind of tool. Not everyone can write a good resume, but everyone has to find a job.

---

• 

  LWilliams369641

  —

  Tried the /interview command. It generates an interview prep package based on archived application materials and can even simulate interviews. Feels like having a personal interview coach. Not as detailed as real human feedback, but as a starting point it exceeded expectations.

---

• 

  贾红_1

  —

  This framework itself might be its best work. Can you imagine the look on an interviewer's face when you say 'I wrote a 19K-star Claude Code job search framework to find a job'? More convincing than any resume.

---

• 

  JaniceTorres_2024

  —

  After using it for three days, the biggest issue is LaTeX fonts. fontawesome5 throws extension errors under MiKTeX, and it took me ages to fix. Switching to a Markdown-to-PDF workflow would be more user-friendly.

No reviews yet