At NVIDIA, I serve as a Tech Lead for AI GPU performance engineering. Key responsibilities include:

• Lead CUDA, Triton, CUTLASS, CuTe, CuTile, and cuDNN kernel optimization for Hopper, Blackwell, and Rubin-class GPUs.
• Improve AI and LLM training/inference by fixing bottlenecks, fusing kernels, and optimizing memory bandwidth.
• Enable PyTorch and JAX integration through XLA, MLIR, TorchInductor, and torch.compile.
• Use Nsight Compute and Nsight Systems for profiling and performance analysis.
• Work on fused GEMMs, attention kernels, distributed training, and inference optimization.
• Drive co-design feedback for future GPU software and hardware.

At AMD, I worked as a GPU performance engineer and technical lead for AI workloads.

• Optimized training and inference on MI300 and MI200 platforms.
• Improved throughput, memory efficiency, and scaling for transformer workloads.
• Collaborated with framework, architecture, and distributed systems teams.
• Contributed to Composable Kernel and ROCm-based AI libraries.
• Optimized attention, GEMM, batching, KV-cache, and multi-GPU communication.
• Helped establish and lead the AMD Center of Excellence in AI at UW Seattle.