Remote job
Member of Technical Staff, Kernels
Job details
About this role
Role overview Design and maintain high-performance compute kernels for model training and inference, with particular attention to the memory use and data movement demands of long-context workloads. You’ll own kernel work through implementation, deployment, and production reliability, balancing speed with correctness and robustness.
Responsibilities - Design and implement kernels that support efficient long-context model workloads. - Own kernel development from initial design through deployment and ongoing reliability. - Improve throughput and latency while maintaining functional correctness and robust behavior. - Test and evaluate whether compute kernels can be ported to alternative hardware. - Work with training, inference, and reinforcement learning teams to shape kernel designs around system needs.
Requirements - Low-level programming experience on AI accelerators, such as NVIDIA Blackwell GPUs or Google TPUs. - Experience developing and optimizing GPU kernels with frameworks such as Triton, CUTLASS, CuTeDSL, Flash-Attention, NCCL, or similar tools. - Deep expertise in computer architecture, low-level optimization, or code generation, with working breadth across machine learning. - Ownership, agility, and persistence in delivering performance-critical systems.
Nice to have - Experience with Pallas, Mosaic, or Mojo kernel authoring frameworks.
Benefits and work setup The role can be based in San Francisco or remotely. The listed US annual salary range is $275,000–$550,000, depending on experience, with equity as a significant part of compensation. Benefits include a 401(k) plan with 6% salary matching, health, dental, and vision coverage for employees and dependents, and unlimited paid time off. Visa sponsorship and relocation support for San Francisco may be available.