Build and optimize an on-device AI inference stack, including custom inference engines, GPU kernels, serving runtimes, model routing, continuous batching, and decoding strategies. Develop C++ systems for low-latency inference across Apple, NVIDIA, AMD, Snapdragon, and other platforms. Profile performance, optimize memory and compute bottlenecks, and translate ML research into production infrastructure.

Lead frontier research in on-device AI, focusing on inference efficiency, model routing, autonomous research pipelines, and real-world evaluations. The role owns significant parts of the research agenda and requires designing experiments, deriving actionable insights, and translating novel ideas into performance improvements. Work may include speculative decoding, quantization, distillation, reinforcement learning, accelerator optimization, and deployment under strict device constraints.