I'm a Senior Staff Software Engineer at Google and a core OpenXLA contributor. My primary focus is scaling XLA:GPU for large-scale pretraining across tens of thousands of NVIDIA GPUs.

• I build the distributed XLA:GPU compiler and runtime infrastructure that keeps pretraining workloads efficient at cluster scale: thunks, command buffers, asynchronous execution, and communication/compute overlap.
• My recent work spans NCCL collectives and symmetric memory, multi-peer synchronization, dynamic slice fusion, and GPU allocator and buffer-assignment improvements.
• I also lead a small team making XLA:CPU work exceptionally well on modern x86 and Arm CPUs at millions-of-QPS scale, and build heterogeneous TPU/GPU + CPU execution for single JAX programs.

_{Scaling XLA:GPU pretraining}

_{Additional compiler work}

_{Earlier work}

_{Generated with github-readme-terminal and refreshed weekly by GitHub Actions.}