• LLM training

• Multimodal and diffusion model inference

• Large model (Gemini, Veo) serving.

• LLM quantization and sparsification.

• TPU robustness and debugability.

• Tensorflow quantization.