Portfolio & Case Studies

AI-powered job application manager that eliminates manual data entry by intelligently extracting details from email using Gemini AI.

An automated blog engine produced 421 articles and 1.27 million words across four sites in five months. Every article passed a quality gate. Every dashboard was green. It earned 10 clicks — and when I joined what we ranked for against what people actually search for, 40% of our page-one keywords had zero monthly search volume. Rebuilt on real demand from DataForSEO and Search Console.

A pre-launch tenant-isolation audit on MarketingSoHigh paired an automated scanner with adversarial verification: 52 cross-tenant leaks confirmed and fixed, 11 refuted. The worst hole never appeared in the scanner's output, because its detector could not parse how one file read its input — and a shrinking finding count looked like progress.

A page in Kompense was summarising an arbitrary 1,000 of 10,911 rows. An agent diagnosed it and wrote a careful fix. Instead of merging it, I had 23 independent reviewer agents attack the change, and every finding had to survive an attempt to disprove it. Nineteen came back, nine survived, two critical — including a fallback that re-created the bug it was fixing. All fixed before merge.

I built a fact-check gate so a fabricated marketing post could never reach me looking clean. When I audited it, the notification reported only the first hold it found, so a post claiming an outage that never happened arrived labelled “nothing is wrong with this post”. The contradiction check failed the other way too, flagging real events as invented. The verification needed verifying.

One React codebase reaches the web, a native Android app on Google Play (Capacitor, with a custom in-app browser in Java) and a Manifest V3 Chrome extension across 103 content-script patterns with five permissions. It reads any job page, scores the fit, writes a tailored CV and cover letter, and fills the form — but never submits for you.