I’m interested in anything that makes training and inference faster, from low-level kernel programming to high-level model architecture design. I also like open source! I’ve made (small) contributions to Huggingface and Triton in the past.

During my time in CAT Lab, I worked on various parts of Triton-viz (multi-DSL support, NKI interpreter, visualizer improvements) and a visualizer for Triton’s linear layouts.