AI explorations

Interfaces for AI-native products — including the thesis I'm currently writing. This page gathers the AI-age work: Plate, the prototypes from various ventures and projects, and the internal tools built from our own pain. Full write-ups are in progress — for now, the shape.

It's easy to mistake the tooling for the work. Every project here started by tearing into some new AI workflow — a whole app spun up in Lovable, design run through Cursor with version control, a particle animation generated by Gemini, a multi-agent pipeline orchestrated with Claude. All of it fun. But strip the novelty away and the job underneath is the same one it's always been: deciding what's worth making, and why. The tools kept changing. What made the work good didn't.

Plate — a GUI for the AI-native device that's coming

A thesis on which interaction patterns survive, which retire, and which still need to be invented. The thinking is recorded; the demo and interaction samples are still underway.

Read the thesis →

Qubitra — a quantum venture's first tangible demo

late 2025 · Fujitsu × Standard Chartered. Built the prototype AI-first, turning an abstract quantum-computing concept into something a stakeholder could see and hold.

PocketSeed — finding the product inside the vision

2026 · claims management. Validated a sweeping vision against real users, extracted the core customers actually paid for, and designed & shipped the v1.

Orion & Kotobase — a bilingual studio, understood

2024–2025 · solving our own pain. An EN speaker embedded in JP-only teams; leaders who couldn't read a room across the language line — so we built the fix, then it spread. Orion brings real-time interpretation into Google Meet, in any language; Kotobase adds a translation layer to Slack, so people write in their own language, JP or EN, and are understood. Both in daily use; Kotobase is now in active pilot.