Background
Research career in text-to-image diffusion at Google Brain, where he was a lead author on Imagen, before co-founding Ideogram in 2022 with Norouzi, Chan and Ho. His LinkedIn headline is the generic 'Research & Engineering'; ZoomInfo still lists him as co-founder at Ideogram.
In a company with a flat member-of-technical-staff structure and no dedicated human-data function, the founding researchers are the people who decide what human judgement gets collected and how it is folded into training. That is the reason to track him despite the thin public trail.
What they run now
Image-model research at Ideogram. Ideogram 4 (Jun 2026) is a from-scratch single-stream diffusion transformer released open-weight; its published evaluation leans heavily on designer judgement, including the Contra Labs blind study with ten professional designers.
What they buy
Inference: post-training preference data and evaluation sets for image models — pairwise designer preference, typography and layout correctness, prompt-alignment sets. Ideogram's published benchmark mix (7Bench for layout control, SpatialGenEval, X-Omni OCR for text rendering, Prism for prompt alignment, plus human studies) shows what his team already measures; the human half is the part a vendor can supply.
Career
- 2022 – presentCo-founder, Ideogram — Current day-to-day role not verified
- 2022Research (Imagen), Google Brain — Lead author on Imagen
