Sayan Shaw

Experience

1. Software Engineer

   Microsoft

   Aug 2022 — Present · Redmond, WA, US

   Led cross functional architectural design, planning, execution and delivery of ML acceleration for DALL-E powered Bing Image Creator’s safety model using ONNX; proliferated daily inference of ONNXRuntime (ORT) by 13.4 million and added 1P partners for ORT in Bing, Office and Windows Silicon organizations.Drove improvement of state-of-the-art NLP and vision models through pre- and post-processing custom operators such as C++ tokenizers for GPT, Llama, Falcon, CLIP (for Stable Diffusion) and RoBERTa. As of early 2024, ONNXRuntime tokenizers are the fastest in the world, beating their respective benchmarks from OpenAI & HuggingFace.Developed a DSL for generating optimized ONNX kernels and writing simplified high performance GPU code on Nvidia V100 and A100 GPUs using TensorCore by improving data movement and compute; collaborated with Microsoft Research teams in Cambridge, UK and Redmond, Washington for AMD GPU support.

Education

• University of California San Diego

  Bachelor of Science - BS

• University of Illinois Urbana-Champaign

  Master of Science - MS