Currently pursuing or recently completed a graduate degree (MS or PhD, or equivalent research experience) in machine learning, applied mathematics, physics, electrical engineering, computer science, or a related field. Publications are a strong plus, but not required at the resident level; a solid research track record is what matters.
A working understanding of large model inference: attention mechanisms, KV cache, long-context decoding, memory bandwidth constraints. Deep production experience isn't expected at the resident level, but the intuition should feel familiar.
Some exposure to inference optimization (quantization, sparsity, kernel fusion, or memory-efficient attention), or to stochastic systems, probabilistic methods, numerical analysis, or analog computation.
Strong Python skills, plus some experience with a systems language (C++, Rust, or similar) or with implementing algorithms close to hardware, beyond high-level frameworks.
Some experience (course projects, research, or otherwise) turning a research idea into working, reasonably reproducible code. You care about whether results replicate, not just whether a demo works once.
Comfort reasoning from first principles about what a novel substrate can do efficiently.
Bonus: exposure to analog or mixed-signal systems, in-memory compute, or non-von-Neumann architectures; publications or open-source work in efficient inference, stochastic algorithms, or novel computing; or experience working on hardware that did not yet exist when you joined.
Strong written and verbal communication skills; prior experience writing up research (papers, theses, technical reports) is a plus, as is any experience presenting technical work to a live audience.
A bias toward ownership and self-direction. You're energized, not overwhelmed, by the ambiguity and speed of a small, fast-moving startup.