Summary
Tianshu Yu is a software engineer with 10 years of experience specializing in scalable reinforcement learning and high-performance inference infrastructure for large multimodal models and autonomous agents. Based in San Jose, he has shipped production-grade systems at ByteDance that scaled GRPO training of a 235B MoE model across 128 H100s, improved MFU by 5×, and introduced async pipeline parallelism and custom inference engines to enable true on-policy RL at trillion-parameter scale. His background includes an MS in Quantum Computing from Duke and a strong foundation in mathematical reasoning, which he leverages to optimize kernels, parallel methods, and hardware bottlenecks. Now at Liquid AI, he continues to focus on infrastructure that advances towards AGI/ASI, blending research-grade rigor with pragmatic systems engineering. Notably, he often combines low-level GPU profiling (NSYS/NCU) with algorithmic changes to unlock order-of-magnitude improvements in throughput and latency.
10 years of coding experience
2 years of employment as a software developer
Master of Science - MS Quantum Computing, Master of Science - MS Quantum Computing at Duke University
Bachelor's degree Mathematics, Bachelor's degree Mathematics at Indiana University Bloomington
Chinese, English