Xingchen Zhao

About

I’m a Machine Learning Engineer at Meta working on ML Infra, inference performance, and generative AI systems. My work spans CUDA/Triton kernels, sparse attention, production model serving, and agentic evaluation pipelines. Previously, I worked on post-training for multimodal LLMs for controllable video generation at TikTok GenAI, real-time perception models at Amazon Robotics, and generative AI products as an early founding engineer at Thinkverse. I’m especially interested in building AI systems that move from research capability to real-world usefulness — faster inference, better post-training, reliable agents, and tools that change how people build software.

Experience

Education