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

What You'll Do5

• Write adversarial tasks that expose where frontier models break on accelerator code, such as Pallas kernels on TPU, and pair each one with a correct, well-structured solution.
• Review ML systems tasks and submitted solutions, then give written technical feedback that holds up under expert scrutiny.
• Build rubrics and guidelines covering kernel optimization, profiler output interpretation, distributed systems reasoning, and serving throughput versus latency trade-offs.
• Advise research and engineering teams on gaps in ML systems, compilers and training infrastructure so model performance improves.
• Work alongside other subject matter experts to keep training data consistent and accurate.

Requirements8

• 2+ years of hands-on professional experience in ML systems, ML infrastructure, model serving, or accelerator performance engineering. This is a systems role, not applied modeling or data science.
• Practical work in at least one of these areas: custom kernels (Pallas, Triton, CUDA, CUTLASS), profiling and trace analysis (JAX/XLA profiler, Kineto, torch.profiler, Nsight), debugging distributed or accelerator-bound workloads, or large-scale LLM serving (vLLM, SGLang, TensorRT-LLM, JetStream, KV cache, paged attention, continuous batching).
• Production experience with JAX, PyTorch, or both.
• Depth at the framework level counts in your favor: custom operators, sharding and distributed training (pjit/shard_map, FSDP, Megatron, MaxText), or compiler and graph-level work (XLA/StableHLO, torch.compile, MLIR).
• Familiarity with modern accelerators such as TPU v4 through v6e, H100, B200, Trainium or MI300, plus the ability to reason about throughput, latency and memory trade-offs.
• A clear record of career progression.
• Availability for at least 40 hours per week on weekdays.
• Strong written communication and the ability to explain complex technical decisions to expert readers.

Who Should Apply

An ML systems or infrastructure engineer who has shipped kernel, profiling, or serving work on real accelerators and wants to move into AI training data. Strong writers who can defend technical choices in review, and who bring TPU or JAX depth, fit best.

Salary Insight