Lead end-to-end model optimization and transformation for large language models, vision language models, diffusion, and multimodal models on Qualcomm inference accelerators. Architect PyTorch-based optimization strategies, drive graph capture and deployment using PyTorch/ONNX/torch.compile, and design fusion kernels using Triton or similar DSLs.… Partner with compiler, performance, and accuracy teams to co-design lowering strategies, optimize transformer-specific patterns (KVcache, decoding, long context), and scale distributed inference across multi-core and multi-device systems. Requires expert-level PyTorch proficiency, deep transformer architecture knowledge, hands-on experience with torch.compile/TorchDynamo, and strong foundation in ML accelerators and distributed systems.