GPU Kernel Evaluation Expert

About Open Train

Open Train AI is the hiring and contracting organization for this role. Open Train is the #1 platform for finding and building careers in AI training and data labeling, helping contributors discover projects, build a professional profile, and apply in minutes.

Creating an Open Train account is free, and this opportunity offers a way to apply specialized GPU programming expertise to the development and evaluation of advanced AI systems.

AI training is the human side of building artificial intelligence. Specialists review code, assess model outputs, and provide structured feedback that helps AI systems become more capable, reliable, and useful.

In this role, your technical evaluations will support training and evaluation workflows involving GPU and accelerator kernel development. The work is fully remote and combines deep engineering expertise with cutting‑edge AI development.

Open Train AI is seeking a GPU Kernel Evaluation Expert to assess the quality, correctness, and completeness of GPU and accelerator kernel development tasks used to train and evaluate frontier AI models.

You will evaluate numerical correctness, benchmarking fairness, task scope, compilation validity, and runtime behavior across a range of kernel task types. Each submission requires clear, rubric‑based written feedback.

• Fully remote contract role for eligible candidates in the United States
• Pay range: $70.00 to $90.00 per hour
• The role description specifies a 40-hour-per-week commitment
• The structured time requirement is listed as 20+ hours per week
• Employment type
  
  s: contractor and part‑time
• Primary language:
  English

You will review technical submissions and task designs across multiple aspects of GPU and accelerator kernel development. Your assessments should be accurate, consistent, and grounded in the applicable evaluation rubric.

• GPU and accelerator kernel tasks for quality, correctness, and completeness
• Numerical correctness using absolute, relative, and ULP tolerances
• Selection and suitability of reference implementations
• Performance‑benchmarking fairness and profiling results
• Compilation and runtime validity across different environments
• Generation from specification, translation or lowering, migration, debugging, optimization, and operator‑fusion tasks
• Written, rubric‑based feedback for every evaluated task

The listing is marked entry level, but the role specifically requires at least three years of hands‑on experience developing, optimizing, or verifying GPU or accelerator kernels. You must have experience in at least two of CUDA, Triton, NKI, or Pallas for JAX.

• 3+ years developing, optimizing, or verifying GPU or accelerator kernels
• Hands‑on experience with at least two of CUDA, Triton, NKI, or Pallas for JAX
• Strong understanding of numerical‑correctness criteria for kernels
• Experience with Nsight, NCU, roofline analysis, or framework‑native profiling tools
• Familiarity with common compilation and runtime failure modes
• Experience with at least three task types: generation, translation or lowering, migration, debugging, optimization, or operator fusion

The following experience is helpful for evaluating a broad range of kernel tasks and accelerator environments. These qualifications are preferred background rather than listed minimum requirements.

• Experience across NVIDIA GPU ecosystems such as CUDA and Triton
• Experience with custom‑accelerator ecosystems such as NKI, Pallas, or TPU
• Compiler engineering, MLIR, or intermediate‑representation lowering
• Memory‑hierarchy optimization, including shared‑memory tiling, register pressure, bank conflicts, and coalescing patterns
• Contributions to cuBLAS, cuDNN, Triton community kernels, or JAX/XLA custom calls

Modern AI systems depend on people who can inspect technical outputs, identify failure modes, and distinguish reliable results from misleading ones. By evaluating kernel implementations and their benchmarks, you help improve the quality of the data and feedback used in AI development.

This is a specialized path within the broader AI‑training industry, where technical contributors can use software and systems expertise to shape how advanced models are built and evaluated.