WHAM (Muse)
World and Human Action Model, a generative model of gameplay trained on Ninja Theory's Bleeding Edge. Open weights (200M and 1.6B) released with the Nature paper, from the Microsoft Research programme I ran.
Senior Technical Programme Manager, Generative AI · Canva
I take generative AI from research to products people use.
I lead generative AI delivery at Canva.
Before Canva, I spent three years with the Game Intelligence team at Microsoft Research Cambridge. There I ran the programme behind WHAM (Muse), a generative model of gameplay published in Nature in 2025, and I’m one of its ten primary authors. I also worked on its real-time successor, WHAMM, which shipped as a playable Quake II demo in Copilot Labs.
Most of my work sits where research meets launch: timelines, data, ethics review, licensing, and the steps it takes to get a model in front of users. Earlier in my career, I led an extended reality team at A*STAR in Singapore and ran national STEM programmes at IMDA.
Outside work, you’ll usually find me on the water. More on that in Beyond Work.
World and Human Action Models towards gameplay ideation
Nature 638, 656-663, 2025
World and Human Action Model, a generative model of gameplay trained on Ninja Theory's Bleeding Edge. Open weights (200M and 1.6B) released with the Nature paper, from the Microsoft Research programme I ran.
Browser simulator for marathon canoe and kayak racing. Reshape the river, model current and boat wash, and race K1/K2/C1/C2 formations to find the faster line before race day.
Standalone MIDI router for Raspberry Pi. Route, remap channels and filter USB MIDI gear from a web dashboard, then play with no computer or DAW.