GPU Kernel Expert

GPU Kernel Expert Remote within the United States | Independent contractor | US$70-US$90 per hour HumanitApp is recruiting for an external hiring partner. This is not employment with HumanitApp. About the work Review GPU and accelerator coding tasks used to train and evaluate AI systems. Your job is to check that kernels produce correct results, that performance comparisons are fair, and that the tasks can compile and run as described. You will write clear feedback against an evaluation rubric. What you will do Check numerical results against suitable reference implementations and error tolerances. Review performance measurements, profiling results and benchmark design. Identify compilation and runtime issues, including memory limits, driver mismatches and incorrect launch configurations. Assess whether task instructions are complete, realistic and technically sound. Explain problems and recommended corrections in concise written feedback. Required experience At least 3 years developing, optimizing or verifying GPU/accelerator kernels. Hands-on work in at least two of CUDA, Triton, NKI or Pallas (JAX). Strong understanding of numerical correctness, including absolute, relative and ULP tolerances. Performance profiling and benchmarking experience with tools such as Nsight, ncu, roofline analysis or framework-native profilers. Familiarity with compilation and runtime failures, such as out-of-memory errors, shape/stride mismatches and autotuning failures. Experience in at least three of these kernel task areas: generation from specifications, translation between frameworks, hardware migration, debugging, performance optimization or operator fusion. Helpful, but not required Work across NVIDIA GPUs and custom accelerators such as TPU or Trainium. Compiler engineering, MLIR or intermediate-representation lowering. Memory optimization involving shared memory, register pressure, bank conflicts or coalesced access. Contributions to kernel libraries or open-source GPU tooling. Terms Independent-contractor engagement, not a W-2 employee position. US-based remote work, completed on your own schedule. Advertised compensation is US$70-US$90 per hour. Final rate, workload and engagement terms are determined by the hiring partner; hours are not guaranteed. Payments are weekly through Stripe or Wise for services rendered. Projects may be extended, shortened or concluded early based on needs and performance. The hiring partner currently cannot support H-1B or STEM OPT candidates for this role. Do not share confidential code or information belonging to a current or former employer or client. How to apply Apply through LinkedIn, or email your CV to [email protected] with the subject "GPU Kernel Expert". Briefly identify the two or more kernel technologies you have used and one relevant profiling or optimization project. Do not send confidential code or identity documents. The external hiring partner determines eligibility, screening, selection and contract terms. Applying through HumanitApp does not guarantee an interview, work or an offer. HumanitApp charges no application fee and may receive referral compensation. Qualified applicants are considered without regard to legally protected characteristics; reasonable accommodations may be requested.