bruce lebizna
i write code in paris, mostly for mistral. these days it is inference work: making the models faster and cheaper to serve.
outside of work i fell into solana three months ago and havent quite climbed out. the rest of the time i play too many video games and pretend that counts as a hobby. :)
work
-
mistral ai, applied llm2024, ongoing
inference, quantization, the unglamorous parts that make models cheap to run.
-
nvidia research, internsummer 2023
poked at h100 memory access patterns. learned why fused kernels are a love language.
-
citadel securities, internsummer 2022
low latency pipelines and backtesting frameworks. first time i cared about cache lines.
school
polytechnique, applied maths + cs, 2020 to 2024. final project on sparse attention.