i build ai systems, mostly the parts nobody demos: agent orchestration, evaluation loops, and the semantic layer underneath that has to be right before any of it works. currently senior ai engineer at perch insights.
before that i co-founded archanan in singapore and ran it as ceo for four years — we built cloud emulators of supercomputers so people could develop at scale without waiting in a queue. then led engineering at musiio, a music-ml company soundcloud later acquired. earlier still: wavelet bases for audio compression at stony brook, and firmware for guitar pedals before that.
i write rust on weekends, mostly audio synthesis and probabilistic programming. the backgrounds on this site are live simulations rather than video — the gear icon opens their controls.
what i work on
consulting
i take on a few engagements a year, split about evenly between two kinds of problem people bring me. the first is a team whose llm prototype demos well and falls over in production — i get called in after that happens, and the fix is almost never the model; it is the data model, the evals, or the failure handling. the second is the earlier question, before anything is built: what to build with ai, or whether to build it at all.