I build robots that live in the physical world.

My background is behavioural science — I came to robotics because what I've always cared about is interaction: what makes something feel alive, safe, and worth talking to. Now I spend my days building exactly that.

• A desktop companion robot — a personality "head" that talks, looks at you, and moves: a real-time voice loop (streaming STT → LLM → TTS, ~2s end-to-end), muscle-aware lip sync rendered on a chest display, head pose that follows the conversation — running across a Mac brain and ESP32-C3 body over USB. The lip-sync engine is being open-sourced as robot-lipsync.
• An underwater exploration robot — early stage, because the ocean is the other physical world worth exploring.

• dont-reinvent-the-wheel — a Claude skill that searches before it builds and before it debugs: external libraries, my own engineering log, and known dead ends, with an explicit reuse / fork / build verdict.
• ai-roundtable — a Claude skill that runs a grounded, turn-based roundtable between Claude and web ChatGPT through the browser, no API keys.

I document everything: engineering logs with symptoms, root causes, fixes, and costs — including the dead ends. Solved problems should become a craft manual, not vapor. Most of what I know about building I learned by shipping one robot, breaking it, and writing down why.