Tingkai Liu 刘廷恺

cooking something up… more to come.

From early 2025 until recently, I worked on fullstack LLMs at Microsoft AI in both IC and tech lead capacities. My responsibilities expanded over two promotions as I contributed to MAI-1-Preview, MAI-Thinking-1, and the next generation of MAI-Thinking models. My work at MAI spanned pretraining (e.g. sparse MoE, long-context extension, vision-language modeling, and multi-token prediction), midtraining data mixtures (e.g. rebuilding MAI’s midtraining effort from the ground up), and post-training recipes (e.g. leading the development of on-policy distillation and landing it in the mainline recipe, as well as scalar and generative reward modeling). I also built core infrastructure from configuration system to job-launch entry point, from custom attention implementations to on-policy distillation loss kernels.

From mid-2024 to early 2025, I was one of three NeuroAI Scholars at Cold Spring Harbor Laboratory, directing research at the intersection of artificial and biological intelligence.

From 2023 to 2024, I was a founding member of ByteDance’s SEED, where I worked on foundation models, multimodal understanding, alignment, and code generation. My work there was recognized with three spot bonuses and a promotion.

I earned my MS/PhD in Electrical Engineering from Columbia University in late 2022, where I studied the computational principles of biological neural systems and was a recipient of the Presidential Distinguished Fellowship. Before Columbia, I earned my BS in Electrical Engineering, magna cum laude in three years, from Rice University.