ML systems researcher with a builder lens

Zhifei Li

Incoming PhD student in Computer Science, Fall 2026

I'm an incoming PhD student in Computer Science starting in Fall 2026. I previously worked as a Visiting Student Researcher at Sky Computing Lab at UC Berkeley, advised by Ion Stoica, and also collaborated with Joseph E. Gonzalez and Matei Zaharia. I completed my B.S. in Computer Science (Turing Honors Class) at Renmin University of China.

I work broadly on machine learning systems, with interests in efficient infrastructure, training and inference systems, and the design of systems for emerging AI workloads.

Research Interests

  • Machine learning systems
  • Training and inference infrastructure
  • Efficient systems for AI workloads
  • Systems support for emerging AI applications

Selected Publications

View all

* denotes equal contribution.

SkyWalker: A Locality-Aware Cross-Region Load Balancer for LLM Inference

Tian Xia, Ziming Mao, Jamison Kerney, Ethan J. Jackson, Zhifei Li, Jiarong Xing, Scott Shenker, Ion Stoica

EuroSys 2026 · Preprint

Paper

SkyNomad: Cost-Effective Multi-Region Scheduling for Deadline-Sensitive Workloads on Spot Instances

Zhifei Li*, Tian Xia*, and others, Ion Stoica

OSDI 2026 · In submission

Paper

LEANN: A Low-Storage Vector Index for Personal Devices

Yichuan Wang, Zhifei Li, Shu Liu, Yongji Wu, Ziming Mao, Yilong Zhao, Xiao Yan, Zhiying Xu, Yang Zhou, Ion Stoica, Sewon Min, Matei Zaharia, Joseph E. Gonzalez

MLSys 2026 · To appear

Paper

Barbarians at the Gate: How AI is Upending Systems Research

Audrey Cheng*, Shu Liu*, Melissa Pan*, Zhifei Li, Bowen Wang, Alex Krentsel, Tian Xia, Mert Cemri, Jongseok Park, Shuo Yang, Jeff Chen, Aditya Desai, Jiarong Xing, Koushik Sen, Matei Zaharia, Ion Stoica

arXiv · 2024

Paper

FrontierCS: The Next Frontier of Computer Science

Qiuyang Mang*, Wenhao Cai*, Zhifei Li*, Huanzhi Mao*, and others, Ion Stoica, Jingbo Shang, Zhuang Liu, Alvin Cheung

arXiv · 2024

Paper

Open Source and Builder Work

Alongside research, I care about translating ML systems ideas into practical infrastructure and widely usable tools. That builder lens is part of the site, but it should not replace the research identity.

SkyPilot

Top 10 contributor; 70+ issues, 50+ PRs merged; 30,000+ lines of code contributed.

Built the High Availability Controller for SkyServe, later adopted by startups including Hypermode.

LEANN

Led research-to-production translation; 40k+ community downloads.

Technical outreach around the project reached 600k+ views.

Latest Insights

All insights