
fleet
A self-improving agent harness: sensors log every human correction, a deterministic metrician scores the runs, and an adversarially judged fix loop edits the harness itself.
A few things I’ve shipped. Each page shows the product, the problem at its center, and how it was built.

A self-improving agent harness: sensors log every human correction, a deterministic metrician scores the runs, and an adversarially judged fix loop edits the harness itself.

An AI-native input method for Chrome: type English anywhere, get natural Japanese inline. Grew into a full streaming translation layer for the browser.

A brand-naming platform with a calibrated scoring engine: a deterministic Go scorer across 12 axes, plus an LLM judge that stays advisory until it proves agreement with human taste.

A private AI journaling app: write a daily entry, get feedback from personas you create, with client-side encryption the business itself cannot read.

An open-source personal alignment engine: a portable AI skill that learns what you actually value through Socratic conversation, then keeps your actions pointed at it.