By the numbers every figure below is grep-checked against real git history, not recalled
How it actually happened first commit dates, not memory
- Sep 1eu-ai-act-scanner begins — a 4-step wizard that maps each risk question to a specific EU AI Act article.
- Sep 2gdpr-compliance-scanner — screens a file for personal data before it gets shared, entirely in the browser.
- Sep 5freight-fraud-taxonomy and freight-risk-atlas — the open pattern library and the assessment engine built on top of it, false positives documented alongside every indicator.
- Sep 6ai-governance-control-room and the portfolio site itself both begin — four frameworks, one evidence model.
- Sep 7scale-of-everything and quantum-galaxy — the science side, built for the same reason as the rest: to be explored, not just looked at.
- Sep 8shrink-signal and the Thekkethundiyil Kudumbayogam family site — not every project has to be about fraud.
- Sep 9risk-os — an operational risk register that runs entirely client-side, no account, no server.
- Sep 10dora-compliance-scanner — 19 cited requirements across ICT risk, incident-reporting and resilience testing.
- Sep 11fraud-watch begins — the biggest single build of the sprint, a 69-phase freight-fraud arcade game.
- Sep 13risk-swarm begins — seven named agents investigate a real repository together, one of them built to attack the others.
- Sep 18risk-replay — reconstructing a decision's lineage after the fact, not just logging that one was made.
- Sep 19policy-audit and ruleshift — an AI-governance evidence auditor and a game, same week.
- Sep 20risk-ring and nightfall-zero — a fraud classifier honest about its own leakage bug, and a character renderer with a name: The Survivor.
- Sep 21shadow-network and hakai-world — a persistent synthetic economy a BYOK AI council investigates without ever seeing the answer key.
- Sep 22Two days before the last day at Amazon, still shipping: SHADOW//NETWORK gets its card on the homepage, and this page gets written — with Aki and Akisa, the way most of it got built.
Where the commits went top 8 by commit count
Counted straight from git rev-list --count HEAD
on every repository, the same afternoon this page was written. No project's number is rounded up.
Why bother writing this down
Most of what a job search asks you to prove — that you can learn fast, ship honestly, and not flinch from a number you don't like — doesn't show up on a resume line. It shows up in 600 commits across 29 repositories in 22 days, in a fraud classifier that reported its own 100% score as a bug instead of a win, in a control room that re-tiers every obligation the moment one honest fact about a system changes.
None of this replaces five years of catching real fraud on real freight lanes. It's the other half: proof that the judgement from those five years survives being written down as something a stranger can run, read, and argue with.
Two days left as I write this. The tools stay up. Back to the main page →
Update: two more shipped after this page went up — reg-search, a BM25 search engine built from scratch over 56 cited AI-governance requirements, and ONE MORE SHIFT, a farewell arcade game over the real fraud-taxonomy data. Both got a second pass afterward too: reg-search now shows the real per-term score behind every result instead of just the final number, and ONE MORE SHIFT now ramps its own difficulty mid-shift and remembers your best score. This count was never going to be the final one — that was the point.