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

$90–120/hr

rate as listed

Workload: 40+ hrs/week
Engagement: Project-based
Location: Worldwide
Field: STEM and Coding

Create and evaluate expert-level ML systems tasks involving accelerator kernels, profiling, distributed debugging and high-throughput LLM serving. The role requires production experience with JAX or PyTorch, modern accelerators and at least 40 weekday hours with no concurrent engagements.

The work

You will design difficult ML systems problems, write reference solutions and assess model-generated answers. Day-to-day topics may include optimizing Pallas, Triton, CUDA or CUTLASS kernels; interpreting profiler traces; debugging distributed accelerator workloads; and reasoning about LLM serving throughput, latency, memory use, KV caches and batching. You will also develop scoring rubrics and provide written feedback that can withstand review by other systems experts.

What the screen emphasizes

The platform is looking for at least two years of hands-on professional work in ML systems, infrastructure, model serving or accelerator performance engineering, rather than applied modelling or data science. Expect follow-up on production use of JAX and/or PyTorch, concrete work with TPU, H100, B200, Trainium or MI300 accelerators, and at least one relevant specialty such as custom kernels, trace analysis, distributed debugging or scaled LLM serving. Framework internals, sharding, custom operators and compiler-level work are valuable but not required in every case.

Practical details

This is a full-time employment position with Cincinnatus LLC, or an appropriate international entity, with potential placement at an AI lab as part of its extended workforce. The commitment is 40 hours per week during weekdays, with no conflicts or other engagements. The listing does not specify whether the work is remote or how much of the collaboration is asynchronous.

Pay band: $90–120/hr

What they look for

At least 2 years of professional ML systems, ML infrastructure, model serving or accelerator performance engineering experience
Production experience with JAX and/or PyTorch
Hands-on expertise in custom kernels, performance profiling, distributed workload debugging or large-scale LLM serving
Experience with TPU, H100, B200, Trainium or MI300 accelerators
Ability to evaluate technical solutions and write defensible rubrics and feedback
Reasoning about throughput, latency and accelerator memory trade-offs

Included with this listing

Interview prep — 10 questions

Matched to what this platform's screening actually measures. Answer each out loud in 60–90 seconds: claim → concrete example → implication.

What it tests: This tests whether your claimed experience is verifiable, genuinely systems-focused and consistent with the requirement for demonstrable career progression.
A strong answer: Anchor the account in a named workload, framework, accelerator environment and operational constraint, then distinguish your decisions from the team's work. Include measurable outcomes and explain how later responsibilities expanded beyond the initial implementation.
Where people lose it: Describing model quality or experimentation while omitting infrastructure ownership, production constraints and the specific components you personally changed.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.

STEM AND CODING on Sidequest

What stem and coding AI work pays

Rates by specialty, the platforms hiring, and every open role in this field.