Home
Mercor

GPU Kernel Expert

  • Pay: $70 to $90 / hour
  • Contract: Contract
  • Who can apply: United States
  • Posted Aug 25, 2026

In short

Get paid 70 to 90 USD per hour to work remotely as a GPU Kernel Expert. You'll review GPU kernel tasks that help train and test a frontier AI lab's models, from home. If you already know your way around CUDA or Triton, this freelance gig could be a strong fit.

What you will do

  • Check that each kernel gives correct numbers, using clear accuracy limits.
  • Judge whether performance benchmarks are fair and set up the right way.
  • Confirm that kernel tasks are well scoped, complete, and can actually compile and run.
  • Write clear, rubric-based feedback that explains every score you give.

What you need to apply

  • 3+ years of hands-on work building, tuning, or verifying GPU kernels in at least two of: CUDA, Triton, NKI, or Pallas (JAX).
  • Solid grasp of numerical correctness rules, like absolute, relative, and ULP tolerances.
  • Real experience profiling and benchmarking with tools like nsight, ncu, or roofline analysis.
  • Hands-on experience with at least three kernel task types, such as debugging, performance optimization, or operator fusion.

Who can apply

This job is open to residents of the US only. It's a freelance contract, and you can work remotely.

A tip from Romowork

We suggest you spell out your kernel experience in your application: name the tools you've used, like CUDA or Triton, and how many years you've spent on them.

Originally published on Mercor

Related jobs