Shan Yang

Researcher · Multi-modal Reasoning

I work on multi-modal reasoning: making vision-language models reason about physics, geometry, and the world they actually see. My focus is on the unsexy half of the stack — audited training data, hard evaluation benchmarks, and RL recipes that survive the jump from text-only chain-of-thought to multi-modal.

Currently Staff Applied Scientist at Adobe Foundry (Adobe's foundation-model research org), post-training foundation models with SFT, DPO, and RL. Previously Tech Lead at Amazon (GenAI Live Action Studio, Amazon Video Search) and Senior Research SDE at Google Research (multi-modal modeling, AIST++). PhD from UNC-Chapel Hill with Prof. Ming C. Lin on learning physical parameters from video.

Currently exploring: multi-modal GRPO and chain-of-thought recipes for VLMs (Physics-o1, in preparation).