GPU Kernel Architect for AI Training & Evaluation

Mercor; Chicago (IL); On-site; USD 120,000 - 180,000; Full time

Job summary

Mercor seeks a professional to evaluate GPU/accelerator kernel development tasks used in frontier AI model training and evaluation. You will assess numerical correctness, performance benchmarking fairness, task scoping, and compilation/runtime validity across diverse kernel types.

The role requires strong experience with CUDA/Triton/NKI/Pallas ecosystems, profilers, and common failure modes. You will deliver rubric-based written feedback to guide improvements.

Qualifications

• 3+ years developing, optimizing, or verifying GPU/kernel code in two or more frameworks.
• Strong understanding of numerical-correctness criteria and reference implementations.
• Experience with performance profiling and benchmarking tools.
• Familiarity with common kernel compilation/runtime failure modes.
• Experience with at least three kernel task types: generation, translation/lowering, migration, debugging, optimization, or fusion.

Responsibilities

• Evaluate kernel development tasks for correctness, performance, and completeness.
• Provide rubric-based written feedback on scope, correctness, and runtime behavior.

Skills

GPU kernels

CUDA

Triton

NKI/Pallas (JAX)

numerical correctness

performance profiling

benchmarking

kernel task types

Tools

Nsight

ncu

roofline analysis

Job description

Mercor seeks a professional to evaluate GPU/accelerator kernel development tasks used in frontier AI model training and evaluation. You will assess numerical correctness, performance benchmarking fairness, task scoping, and compilation/runtime validity across diverse kernel types.

The role requires strong experience with CUDA/Triton/NKI/Pallas ecosystems, profilers, and common failure modes. You will deliver rubric-based written feedback to guide improvements.