About / Austin Xu
Build the
instrument panel.
I work at the seam between research and the systems that make research observable: kernels, agents, interfaces, and the small details that decide whether a result can travel.
Three recurring questions
Can the path be faster? I like the layer where an abstract model meets memory traffic, a kernel, or a consumer GPU that has to do more with less.
What is hidden? Agents need to reason about actions that are not fully observable. The interface should make that uncertainty legible, not decorate it away.
What changed? A volatility surface is useful when it reveals a regime shift. A benchmark is useful when it keeps the failed comparison in the frame.
Now
Researcher at StarAI, UCLA. Currently circling GPU inference, strategic deception, and uncertain systems.