AI Perf and Optimization@ Google | 2x Google, Moodys

I focus on the space where model performance, training efficiency, and product strategy come together. My work is centered on helping large models train, tune, and scale more effectively by understanding how architecture, data movement, parallelism choices, kernels, and hardware behavior shape real results. I work closely with researchers, model engineers, compiler and runtime teams, and distributed systems groups to identify the factors that influence speed, stability, and overall capability. Earlier in my career, I drove strategy for an AI compiler platform that significantly improved the speed and cost efficiency of large model training and inference. That experience taught me how model structure, execution planning, and system behavior come together to create performance limits. Today, my focus is on the model side rather than infrastructure. I spend my time understanding performance bottlenecks, shaping evaluation and benchmarking methods, guiding training and tuning approaches, and supporting research teams in their ability to experiment quickly and reliably. I am motivated by the challenge of advancing model capability through better training dynamics, smarter performance work, and clear product direction that supports scientific goals. My aim is to help teams building advanced models push the boundaries of what is possible while providing the clarity, tools, and insights needed for fast and predictable iteration.