-
feb 2025 - present
inference optimizer @ avian.io
this required many completely custom kernels for optimizing expert selection + routing, fully custom gemm kernels with a fully custom quantization spec to minimize quantization error, and lots more
-
oct 2024 - feb 2025
member of technical staff @ espresso.ai
trained lots of transformer models on rather unique modalities
-
apr 2021 - oct 2024
cofounder & ceo @ forefront.ai
built the first platform that enabled fine-tuning and deploying of archaic llms like gpt-j, including pioneering multi-lora serving that oss didn't replicate for about a year, achieving $131k monthly peak revenue
created a chat interface with innovative features (first to implement chat sharing) that grew to 3m users
-
jul 2020 - mar 2021
head of ml @ fion
developed sota wildfire spread prediction models using vit unets for the california and colorado state wildfire services
-
jan 2020 - jun 2020
cofounder @ owner.com