About Me

I am a fourth-year PhD student in the Department of Electrical and Computer Engineering at Carnegie Mellon University, advised by Prof. Rashmi Vinayak. I expect to complete my PhD in May 2027. I am a member of the TheSys Group, the Catalyst Group, and the Parallel Data Lab. My research interests span large-scale systems for LLMs and training methods for building more capable models.

I received my bachelor’s degree from the Yao Class (Special Pilot CS Class) at Tsinghua University. During my undergraduate studies, I had the privilege of working with Prof. Yi Wu and Prof. Yang Gao on reinforcement learning and its applications to systems.

Research

My research spans LLMs and large-scale distributed systems. My systems work focuses on improving the reliability and efficiency of large-scale LLM workloads. Current topics include silent data corruption in hyperscale LLM training clusters, in collaboration with Meta, and efficient LLM serving on heterogeneous hardware.

Building on this systems background, I am expanding my research toward the learning aspects of LLM training, with an initial focus on RL post-training. More broadly, I am interested in understanding how training methods and scale shape model capabilities, and using these insights to improve training.

Publications

• OpWeave: Flexible Operator Disaggregation for Heterogeneous LLM Serving

  Zikun Li*, Yixuan Mei*, Shiqi Pan, Zixuan Chen, Xiaowen Zhang, Mengdi Wu, Shuhuai Lin, Yutong Yang, Zhihao Zhang, Xupeng Miao, K. V. Rashmi, Zhihao Jia

  Preprint Paper

• Coral: Cost-Efficient Multi-LLM Serving over Heterogeneous Cloud GPUs

  Yixuan Mei, Zikun Li, Zixuan Chen, Shiqi Pan, Mengdi Wu, Xupeng Miao, Zhihao Jia, KV Rashmi

  Preprint Paper

• SEVI: Silent Data Corruption of Vector Instructions in Hyper-Scale Datacenters

  Yixuan Mei, Shreya Varshini, Harish Dixit, Sriram Sankar, KV Rashmi

  2026 Paper Code · Pending Meta legal approval

• Helix: Serving Large Language Models over Heterogeneous GPUs and Network via Max-Flow

  Yixuan Mei, Yonghao Zhuang, Xupeng Miao, Juncheng Yang, Zhihao Jia, Rashmi Vinayak

  2025 Paper Code

• Quarl: A Learning-Based Quantum Circuit Optimizer

  Zikun Li, Jinjun Peng, Yixuan Mei, Sina Lin, Yi Wu, Oded Padon, Zhihao Jia

  OOPSLA 2024 Paper Code

• SpeedyZero: Mastering Atari with Limited Data and Time

  Yixuan Mei*, Jiaxuan Gao*, Weirui Ye, Shaohuai Liu, Yang Gao, Yi Wu

  2023 Paper Project

* Equal contribution.

Experience

Google Cloud — ML Training (JAX Training) — Software Engineering Intern

May 2026 – Aug. 2026 · Kirkland, WA

Research on training stability and convergence in asynchronous agentic RL post-training, using MaxText, Tunix, and TPU Inference.

Received a Google Spot Bonus from the Director of Machine Learning for enabling Qwen 3.5 RL post-training and achieving over 3× speedup in wall-clock training step time.

Google Cloud — Core ML (GPU Performance) — Software Engineering Intern / Student Researcher

May 2025 – Dec. 2025 · Sunnyvale, CA / Pittsburgh, PA

Research on scheduling algorithms to improve resource utilization for long-context, deep-research generation in large-scale production LLMs.

Teaching

• Teaching Assistant — 15-750: Algorithms in the Real World · Fall 2024