Shan Yang

Staff Applied Scientist, Adobe Foundry | Ex-Tech Lead Amazon | Xoogler

I am a Staff Applied Scientist at Adobe Foundry, where I post-train Adobe's family of generative AI foundation models for customer-facing products powering Firefly and creative tooling. My research centers on four threads: post-training (SFT, RLHF/DPO, distillation, test-time training) to align foundation models with creator intent; multi-modal understanding across video, image, and language; world modeling — building models with internal representations of how the physical world behaves; and reasoning — RL and test-time compute for physics and scientific problem solving.

Previously I was a Tech Lead at Amazon, where I drove GenAI Live Action Studio and Amazon Video Search, and a Senior Research SDE at Google Research working on multi-modal generative models. I completed my PhD at UNC-CH under Prof. Ming C. Lin.