About the role We're looking for an ML engineer who works across the full stack from model to silicon — comfortable optimizing training and inference performance on GPU/AI-accelerator infrastructure, building or tuning models, and adapting model and inference-engine design to the constraints of the underlying chip and its NPU. You'll move fluidly between algorithm work, systems-level software, and infrastructure work, closing the loop end-to-end rather than owning just one layer of the stack. This is a rare chance to work the full cycle of AI silicon, from model down to chip — something most ML engineers at large companies never get access to. What you'll do Optimize training and inference performance across GPU and AI-accelerator infrastructure, including MLOps pipelines Design, train, and evaluate ML models (deep learning, LLM, CV, or recommendation systems) and take them into production Harden and extend NPU cores (e.g. building on an open RVV/tensor core like CoralNPU) into production silicon Build or optimize inference engines and serving runtimes against real hardware constraints — latency, memory, and power Work below the application layer where needed — BMC firmware, embedded Linux, or RTOS (e.g. Zephyr) — so AI features run reliably on real systems Build automated test/verification harnesses that close the loop for AI-assisted RTL/DV, hardware bring-up, or manufacturing test Apply ML to security — AI-driven log/intrusion analysis, AI-assisted penetration testing, or firmware/hardware security work Collaborate closely with RTL/hardware, firmware, and QA teams to ship AI features end-to-end, from training through deployment and monitoring