Yuxuan Zhang
I am an incoming Ph.D. student at UC San Diego HDSI, advised by Prof. Hao Zhang.
I am a Software Engineering Intern at Google, starting June 15, 2026.
I work on agentic inference and reinforcement learning for language models.
Email: yuz165 [at] ucsd [dot] edu / GitHub / X / LinkedIn / Google Scholar
About
My work focuses on building useful AI agents and scalable systems for LLM reinforcement learning, especially rollout infrastructure, digital environments, and agent evaluation.
Previously, I received my B.S. in Data Science from UC San Diego and worked with Prof. Hao Zhang and Prof. Haojian Jin on AI agents and LLM reinforcement learning.
Recent News
Selected Work
NanoRollout
Scalable rollout infrastructure for agentic inference and LLM reinforcement learning.
Ace Attorney AI Gameplay Demo
AI agent playing Ace Attorney, demonstrating game-grounded agent evaluation.
Publications
LMGame-Bench: How Good are LLMs at Playing Games?
Introduces **LMGame-Bench**, a unified Gym-style benchmark that tests LLM agents across platformer, puzzle, and narrative games, addressing vision brittleness, prompt variance, and data contamination.
General Modular Harness for LLM Agents in Multi-Turn Gaming Environments
Introduces a perception-memory-reasoning harness that consistently boosts LLM/VLM gameplay across classic and modern game suites, uncovering module-specific performance patterns.