Most of what I build starts as a favour that got out of hand. Job Scout is the cleanest example I have.
It began with one person — a sharp, accomplished person — who had applied to dozens of jobs and heard nothing back. Not rejections. Silence. The kind that makes you wonder if the "Submit" button does anything at all. I read her résumé and it was fine. Then I read the job postings, then I read up on how the screening software actually parses a document, and the problem snapped into focus: her experience qualified her, but the machine in the middle couldn't read it. She was getting filtered out before a human ever looked.
The first version was just me, reading carefully
The earliest "system" was embarrassingly manual. I'd take a posting, pull out what it was really asking for, and rewrite the résumé so the right language sat where an applicant tracking system expects to find it — without lying, without keyword-stuffing, without turning a human being into a bag of buzzwords. It worked. She started getting calls. And then her friends wanted it, and their friends did, and I realised I had a process I was running by hand over and over.
Anything you do by hand the same way fifteen times is begging to be turned into a system. That's the moment a favour becomes a product. That product became Job Scout.
What the AI actually does (and what it deliberately doesn't)
Here's the part I care about, because it's where AI earns its keep instead of just sounding impressive. Job Scout reads job postings — a lot of them — against a candidate's actual work history. It infers what a role is really hiring for, scores the fit, and tailors a résumé and cover letter to that specific posting so the screening software parses it correctly and a recruiter searching for those skills can find it. Every posting that goes to a client is checked to confirm it's still open, because nothing burns out a job seeker faster than pouring an evening into a role that was filled three weeks ago.
Just as important is what it refuses to do. It does not click "apply" for you. It does not invent experience you don't have. It does not promise you a job — be suspicious of anyone who does. The human stays in the driver's seat; the AI does the reading, matching, and formatting that no person has the hours to do at scale. If you want the unglamorous mechanics — how a résumé gets past an ATS, what a tailored application looks like, how the city-by-city job market guides come together — it's all out in the open on the site.
What building it taught me about AI
Three things stuck with me, and they apply far beyond job hunting.
The leverage is in reading, not writing. Everyone wants AI to generate. The real value in Job Scout is comprehension at volume — reading thousands of postings and one person's history and finding the genuine overlap. Generation is the easy last step.
Constraints are the product. Deciding what the AI won't do — won't apply for you, won't fabricate, won't promise outcomes — is what makes it trustworthy. The guardrails aren't limitations bolted on afterward; they're the whole reason anyone should hand it something as personal as their career.
A favour is the best product spec there is. I didn't start with a market analysis. I started with one person staring at a wall of silence. Build the thing that fixes a real problem for a real person, and the system that scales it tends to reveal itself.
Job Scout is veteran-owned and it's still growing — more cities, more sources, sharper matching every month. But the core is the same as that first afternoon with one frustrated friend: your experience already qualifies you. The job is making sure the machines, and then the humans, can actually read it. If that's the wall you're staring at, go take a look.