Leading ML GPU optimization effort.

2019 - 2022: ML at Google brain. Working on Tensorflow, Pytorch performance on GPUs. Improving ML models like NCF, BERT, resnet50. Official Bert GPU submission to MLPerf / MLCommons. Also wrote some CUDA kernels for optimizations.

2022 - present: Ads Machine Learning (Youtube): Improving model quality to drive conversions and revenue for YouTube Ads. Use new data sources, engineer features, leverage LLMs, and architect current recommendation models to represent users more accurately.

GPU modeling.

Worked on novel techniques to estimate performance and power of GPU/CPUs by using analytical models and architectural simulators.

Worked with the Xeon Phi group. Added a key feature to the performance model to better utilize the memory system.