Zichen Zhao (Danny) is an undergraduate researcher in Artificial Intelligence at the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in Abu Dhabi. He works on LLM personalization, agent memory, AI for science, and building software that puts a good education within reach of anyone who wants one.
Most of his research asks one question in different disguises — how does a system figure out who it is for, and what to keep about them? That question produced PURPLE (ACL 2025), which casts LLM personalization as a contextual bandit problem, and SurMem, where a learned surrogate scores whole memory subsets by predicted downstream reward rather than trusting top-k embedding similarity. He is now pushing that framing toward personalized retrieval as contextual offline black-box optimization, alongside work on discrete diffusion and biomedical foundation models with Prof. Xue (Steve) Liu’s group at MBZUAI and as a research engineer at GenBio AI.
Before MBZUAI he studied at Shanghai Pinghe School, where he shipped PH Clubs, an iOS club-management app for his school, and CareLink, a marketplace for assistive technology. He is now a co-founder of Erudier, which builds a local-first desktop app for personalized learning: students who get one-on-one tutoring outperform their classroom peers by roughly two standard deviations, and Erudier is a bet that software can close most of that gap. Both halves of the work come down to the same problem — adapting to one person instead of the average of everyone.