Transitioned to the management track to better support the growing compiler team.

I lead the ML compiler team at Cruise.Cruise’s autonomous vehicles (AVs) use machine learning to accomplish various critical tasks. These models have to run under a tight latency budget on the resource-constrained compute hardware available on the AV.We ship an ML compiler based on MLIR that compiles PyTorch for efficient execution on the AV. We support NVIDIA GPUs and Cruise-custom ML accelerators.Since I joined in early 2022 I have: - Built the team charter, software architecture & roadmap from ground up. - Supported & led the team as they built & shipped v1.0 of the compiler. - Evangelized this new technology to various customer teams and fostered adoption.

I managed the Google ML GPU team. We were responsible for the GPU backends in TensorFlow & XLA.I developed & owned owned the overall technical strategy for TensorFlow & XLA on GPUs. I liaised with GPU vendors (NVIDIA and AMD) and ensured that Google’s ML software excelled on their hardware.

I was an individual contributor on the XLA team and implemented various performance improvements and features for XLA. Much of my work is open source: https://git.io/JMeb2

LLVM commit history: https://github.com/llvm/llvm-project/commits/main?author=sanjoy

I worked on Falcon, Azul's optimizing JVM JIT compiler based on LLVM. I worked on the project from conception to GA (which took four years end to end) and was instrumental in its success.