Principal Applied Scientist at Amazon, working on hard open problems in multimodal AI: perception, vision-language-action, and world models. For two decades I've taken products from zero to one: advancing the science and shipping products.
The through-line is 0→1: take a research idea and make it survive contact with a real customer.
A 0→1 effort that helped bring few-shot learning into a public AWS service, giving customers a way to build custom vision models without massive labeled datasets. A research direction turned into a product used at cloud scale.
Shaped the multimodal research direction at SambaNova: vision-language and generative models on a new class of AI hardware. The interesting part was which modeling ideas a different compute paradigm makes worth trying.
Brought foundation-model thinking into autonomous-vehicle perception. Investigated the role of foundational models in generalization and in simulation. The real shift was the representations themselves: they change which long-tail problems are reachable at all.