about
quick facts
who: jay kruer.
what (i am doing): building a deep learning framework and training LLMs at tenstorrent. we make chips that make deep learning go brrrr on an open source sw stack at better perf/$ than our competitors. here's a brief explanation gesturing at how we're able to do that: if you're familiar with TPUs, you can think of it as what might happen if you took a TPU and exploded its humongous SRAM and matmul engine over a sea of tiny tiles which can talk to each other. many tiny SRAM strong! 🦧
when: now through ??? you tell me
where: bend, oregon.
with whomst: my wonderful fiancé jaclyn and our dog zuzu. see outsta for some pics of us in the wild :)
before
i've worked as a formal verification engineer at tenstorrent and intel, and before that as a research engineer at sifive.
i studied math/cs at reed college, where i wrote my thesis on the category-theoretic semantics of lambda calculus, specifically as applied to the open normalization problem for the simply-typed lambda calculus. it's pretty wild stuff, and i went to great pains to make it approachable to someone with a cursory background in programming language theory. check it out!
check out my resume for more.