About
I'm Nathan Ku. I'm interested in what makes a solution better, what it costs, and where it breaks. I explore these questions by building things and testing them.
Most of the work on this site started with a concrete constraint—how much computation can be removed without changing a speech feature, how much compute a facility can safely keep running when cooling fails, why a memory layout suddenly stalls a cache, or when a simulation stops agreeing with physics.
I use software, math, and physical models as tools to test those trade-offs rather than as identities. Whenever an experiment reveals a limit or disproves an earlier assumption, I try to leave that boundary visible.
If you're hiring, collaborating, or curious about any of the experiments here: