Muyang (Sandy) Zheng

Muyang (Sandy) Zheng

MSCS Student at UC Davis

About Sandy

I am a second-year M.S. student in Computer Science at University of California, Davis, advised by Dr. Lifu Huang. My research focuses on agentic AI, multimodal reasoning, and trustworthy LLMs.

Previously, I earned my B.Eng. in Information Security at Hefei University of Technology in 2025, advised by Dr. Yuanzhi Yao.

Selected Research Projects

Overview of the GliDe framework for detecting and temporally grounding video game glitches.
Agentic AI · Multimodal Reasoning ACM MM 2026 First author

Open-Ended Video Game Glitch Detection with Agentic Reasoning and Temporal Grounding

TL;DR An agentic framework that detects glitches in gameplay videos, explains what went wrong, and pinpoints when it happened.

Overview of MIST's iterative semantic tuning method for evaluating black-box LLM safety.
LLM Safety · Jailbreak Attack arXiv 2025 (Under Review) First author

MIST: Jailbreaking Black-box Large Language Models via Iterative Semantic Tuning

TL;DR Iterative, meaning-preserving prompt refinement exposes safety vulnerabilities in black-box LLMs without access to their internals.

Long-Horizon Video Reasoning Manuscript in Progress Project Lead

Benchmarking Long-Horizon State Reasoning in Videos

TL;DR A comprehensive benchmark evaluating how multimodal models track entities, reconstruct evolving states, and integrate evidence across long videos.

    News

    • Our work on open-ended video game glitch detection has been accepted to ACM MM 2026! Many thanks to my awesome mentors and collaborators!

    • MIST, our work on jailbreaking black-box LLMs through iterative semantic tuning, is available on arXiv.