CV

Portable expert-parallel communication library for MoE models (1200+ stars)

UCCL-EP

Supervisor: Yang Zhou

• Achieved 2.1x throughput improvement on AWS EFA and 45% higher training throughput for DeepSeek-V3 on AMD clusters. Integrated into vLLM and Megatron-LM.

Benchmarking framework for SmartNICs and DPUs

dpBento

Supervisor: Qizhen Zhang & Philip A. Bernstein

• Designed and implemented a flexible benchmarking framework for evaluating DPU performance in disaggregated environments.

Retry policy analysis in Lotus framework

Distributed Transaction Protocols

Supervisor: Dixin Tang

• Developed multiple retry policies and conducted empirical comparisons across various transaction protocols.

GPU-accelerated SQL engine (800+ stars)

Sirius

• Built hybrid Top-N execution engine, achieving up to 6.12x end-to-end speedup on ClickBench.

TA, Intro to Big Data Systems

University of Wisconsin-Madison

• Fall 2025 & Spring 2026

Ph.D.

University of Wisconsin-Madison

Computer Science

• Advisor: Ming Liu

B.S.

University of Toronto (St. George Campus)

Computer Science

• Advisors: Qizhen Zhang and Phil Bernstein

UCCL-EP: Portable Expert-Parallel Communication

OSDI 2026

dpBento: Benchmarking DPUs for Data Processing

arXiv preprint arXiv:2504.05536