Hi, I’m Kapil. I work as an AI Systems Engineer at Meta. Currently, I work on the Pytorch team and a contributor/maintainer of PyTorch distributed and torchcomms. I also support Meta’s large scale pre-training and mid-training pipelines. Previously, I focused on large scale data preprocessing for recsys via torch.data, torcharrow, etc.

This blog mostly captures my hobby projects and things I experiment with outside of work as and when I get time. My current focus is GPU kernel programming, LLMs, inference performance and comms/compute overlap.

Previously, I have also worked as a Quant at Option Market Making and High Frequency Trading companies, where I worked on low-latency trading strategies and market-microstructure ML modelling.