AI that does
real work.

Replen is a supply-chain planning skill that sits between a general AI and a real operating decision. It sizes safety stock and reorder points, diagnoses forecast error, handles intermittent demand, compares EOQ against MOQ and pack constraints, and segments a portfolio with ABC–XYZ. The more important part is the gate before the number: which service-level definition do you mean, is the variability measure valid, and what data is still missing?

I built it by turning repeated planning decisions into a written answer protocol, moving the arithmetic into five dependency-free Python calculators, adding reusable CSV inputs, and separating public planning knowledge from private work context with a red-line check. Its operating rule: numbers without assumptions are forbidden. It is a live pilot for decision support — no ERP write-back and no invented data.

An org chart for a single life: a chief-of-staff agent on top and specialists underneath — investing research, personal brand, health, the day-job planning stack — plus a horizontal layer that digests everything I read and hear and routes the signal to the cell that should change because of it. Every agent is built the same four ways: a written-down brain, a callable skill, a ledger, a rhythm.

Every agent drafts; one human reviews and decides. No trade is ever executed, nothing is ever auto-posted, and a failed run says so instead of inventing a number. It is how this site, the research notes and a good share of my working week actually get produced — this page included.

A match-logging and hitting-partner app for club tennis players: find someone at your level, set up a hit, chat before the match, submit the score set by set, watch the season build. Designed, built and shipped to production by one person — a supply chain planner, not a software engineer.

The method matters more than the stack. I drew every screen before any code existed — 34 frames and an acceptance spec — then directed AI to write the React and SQL while I made every product call. Five phases went from that design to v1.0 in eleven days, with friends signing up mid-build, and the spec’s fourteen acceptance criteria were walked at the end: thirteen pass, one deferred on purpose. The milestone that counts: a real invite, accepted, a Saturday morning on court, and the score recorded in the app — a loss, logged honestly.