Laid off in March, I built a job-search tool almost entirely with Claude. Here's the whole process.

Back in March I got laid off for the first time in my life. After a short stretch of dread, then acceptance, I started job hunting, and I ended up building a tool for it with Claude that turned into a real product. Here's how it came together. I started by using Claude to sort out my resume. I dumped a long text file of everything from my career into it (projects, technologies, roles, etc.) and had it organize everything into a master resume and act as a career advisor, asking me questions to flesh out my experience. It suggested titles and roles I hadn't even considered but that matched my background. Then I got sick of manually searching LinkedIn, Indeed, and everywhere else for listings. My whole career is in financial markets, so most of what turned up was noise. I had Claude build me a script that scraped a bunch of fintech company ATS boards and dumped the jobs into a CSV. Still too much noise, so next I had it build a scoring system to rank jobs by how well they matched my profile. That grew into separate scores for titles and descriptions, plus a negative keyword system to push down roles I wasn't interested in. Before long Claude had it wrapped in a GUI where I could tweak the keywords and weights and tune the scoring on the fly. Next I wanted to fix tracking applications, which had turned into a spreadsheet mess. Claude added an application CRM with kanban-style tracking. I mark that I applied to something in my matches and it spins up a card I can fill in with the resume I used and any notes. Then I added interview prep and tracking, and eventually a Gmail sync so it updated itself. After a bunch of iterations I did a full refactor. The first version was a Streamlit app on SQLite, which was great for moving fast but hit its limits once I wanted it to feel like a real product with multiple users. The backend stayed Python/FastAPI, but Claude rebuilt the frontend in React/TypeScript and moved the database to Postgres. It runs across Cloudflare, Railway, and Hostinger now. When Fable came out my whole workflow changed. It did a big overhaul of the UI to make it more user-friendly, and after that I mostly used it to research and plan enhancements, then had it spin up subagents to implement the features and review the work before I deployed. Going from "ask AI to write me a script" to "Fable plans it, subagents build it, Fable reviews it" was the biggest jump in how much I could get done. Somewhere in there I added an analytics section to figure out what was actually working: reply rates by score band, which resume version performed best, whether referrals helped, whether applying within 48 hours of a posting mattered. It also tracks which keywords on the jobs I applied to correlated with reaching a screen or an interview. I landed a new role after 9 weeks and have kept building on it since. What it is: sort of a cross between Hiring.Cafe and Huntr, except the jobs get scored on your keywords and served up to you instead of you running the same searches over and over. I tried both of those before building my own and neither quite did what I wanted. It's free to try, with an optional paid tier. Link is searchsteward.com. Happy to get into the process, the stack, or the Claude/Fable workflow in the comments if anyone's curious.