Shengzhe Zhang

M.S. Computer Science at UC Irvine, graduating December 2026. Looking for MLE or SWE roles in generative AI and NLP — Bay Area preferred.

I'm Shengzhe Zhang, or Sean Zhang. I go by BaiShiXi online. I'm finishing an M.S. in Computer Science at UC Irvine in December 2026, after a B.S. in CS at UC Davis.

What I'm building

We Snow — a ski trip-planning app I co-founded, now with several hundred registered users on iOS and Android. Pick a mountain, get a group together, sort out carpools, and split the costs afterwards without anyone doing arithmetic on a napkin.

Most of my work on it is the language-model side: filling gaps in the resort database with LLM pipelines, building fuzzy semantic search over it with RAG, and turning live resort weather telemetry into personalised travel and gear recommendations. The other half is less glamorous and just as satisfying — an automated multi-party expense split, and a database schema redesign that cut authentication latency.

What I'm researching

I'm a graduate research assistant in human-computer interaction at UC Irvine, looking at how people actually work with AI-assisted systems — which turns out to be a different question from how good the system is on its own. Still in progress, so there isn't much to link yet.

Before that, robot learning: how much of a vision-language-action model's competence lives in the backbone versus the action head, and what breaks when you shrink either one.

Recent writing

Variational Autoencoders in Topic Modeling — reimplementing AVITM in PyTorch and checking whether a learned inference network really does match classical topic-model inference at a fraction of the cost. It does: an 88% cut in training time at competitive perplexity and coherence.

How Much Policy Does a VLA Need? — scaling only a VLA's diffusion policy head to find out how much capacity the action policy really needs.

Human-Like Fox v Fox Bots for Melee — whether reinforcement learning started from professional demonstrations beats the same pipeline started from amateur ones.

More on the projects page.

Looking for work

I'm looking for machine learning engineer or software engineer roles starting after I graduate in December 2026, with a preference for generative AI and NLP. I'd like to be in the Bay Area — San Jose especially — and I'm currently in Irvine.

Closest to what I want to do more of: LLM pipelines, RAG and semantic search, and the evaluation work that tells you whether any of it actually helped.

The fastest way to reach me is email.

Elsewhere