ResumeTailor

I am user zero. This is not a portfolio piece that lives in a slide. It's the system I actually run my job hunt on, and being open about that is the point: it shows what I build for my own leverage when nobody assigns me the work.

The product decision that matters

The first version tailored everything I fed it. That was wrong: the scarce resource in a job hunt isn't tailoring capacity, it's judgment about where to spend effort. So the pipeline now leads with a fit-first triage gate: a 0-100 score, a verdict, and two or three one-line reasons, in about thirty seconds. Most job descriptions should die there. Only roles that clear the bar get the full run: tailored resume PDF with the same layout as my master resume, a changelog of every edit, cover letter, outreach messages, application answers, and interview prep notes.

Honesty as architecture

A tailoring system's failure mode is drift into fiction. ResumeTailor logs every change it makes to my resume in a per-application changelog, and anything that stretches beyond established fact is flagged and requires my explicit per-item confirmation before it ships. I designed the system so it can't quietly embellish. The constraint lives in the pipeline, not in my discipline.

What broke along the way

The Chrome extension's autofill hit IPv6-only ATS endpoints, the eval of "good tailoring" turned out to be much harder than generating it, and early triage verdicts were confidently wrong until the scoring rubric got grounded in my actual profile files rather than the model's general taste. The funnel analytics view exists because I kept lying to myself about my own pipeline. Revealed preference beats stated preference, even when both are mine.