I am a Ph.D. student in EECS at UC Berkeley, advised by Ion Stoica. I work on full-stack machine learning systems, from kernel optimization and efficient system design to text and multimodal algorithms, with the goal of making modern AI workloads efficient on real hardware.
I am a member of Sky Computing Lab and LMSYS. My work spans the full stack of machine learning systems: kernel optimization at the hardware-software boundary, efficient system design for large-scale inference and generation, and text and multimodal algorithms that benefit from those systems advances.
I am especially interested in algorithm-system co-design: building methods that are not only theoretically appealing, but also practical and efficient when deployed at scale. Recently, I built FreeToken, an edge-native MoE serving engine that runs 290B+ frontier models on consumer GPUs, and FlashLib, a GPU library that turns classical ML operators such as K-Means, KNN, PCA, and SVD into fast online primitives. Earlier work covers LLM serving, sparse attention, and efficient video generation.
Recent highlights include the Amazon AI PhD Fellowship, a research scientist internship at Amazon Neuron Science, and a research internship at Meta.
Previously, I graduated from the ACM Honors Class at Shanghai Jiao Tong University.
