MLOps Engineer, LLM Performance (Kernels, Profiling, Serving)

A full-time remote role for ML systems engineers who live in kernels, profilers and serving stacks. You design hard tasks on accelerator code, profiling, distributed debugging and LLM inference serving, write strong solutions, and build the rubrics that judge them. Open worldwide, and JAX, XLA or TPU experience is especially valuable.

What you will do

• Design challenging tasks across kernels, profiling, debugging and inference serving, including adversarial problems that show where frontier models fail on accelerator code (for example Pallas kernels on TPU).
• Write accurate, well-structured solutions to those tasks.
• Review ML systems tasks and solutions, with clear written technical feedback that stands up to expert review.
• Write guidelines and detailed rubrics on kernel optimization, profiler output, distributed systems reasoning, and serving throughput and latency trade-offs.
• Help research and engineering teams close knowledge gaps on ML systems, compilers and training infrastructure, and work with other experts to keep the work consistent.

What you need

• 2+ years of hands-on professional experience in ML systems, ML infrastructure, model serving or accelerator performance engineering. This is a hands-on systems role, not applied modelling or data science.
• Practical experience in at least one of: custom kernels (Pallas, Triton, CUDA, CUTLASS); profiling and trace analysis (JAX/XLA profiler, Kineto, torch.profiler, Nsight); debugging distributed or accelerator-bound workloads; serving LLMs at scale (vLLM, SGLang, TensorRT-LLM, JetStream, KV cache, paged attention, continuous batching). More than one is a strong plus.
• Production experience with JAX and/or PyTorch. Framework depth is a strong plus: custom operators, sharding and distributed training (pjit/shard_map, FSDP, Megatron, MaxText), or compiler work (XLA/StableHLO, torch.compile, MLIR).
• Experience with modern accelerators such as TPU (v4 to v6e), H100, B200, Trainium or MI300, and the ability to reason about throughput, latency and memory trade-offs.
• Demonstrable career progression.
• Reliable availability for at least 40 hours a week on weekdays.
• Strong written communication and the ability to explain complex technical decisions clearly.

Good to know

• This is a 40-hour full-time engagement with no other engagements alongside it.
• Depending on where you live, you are engaged through a staffing company or one of its local employer-of-record partners, with the chance to be placed in an AI lab's extended workforce.

How pay works

This job is paid hourly, at the rate listed above, for 40 hours a week.