Remote GPU Kernel Auditor for AI Tasks

Jobgether; South Africa; On-site; ZAR 1,539,000 - 2,198,000; Full time

Benefits offered by this job

Competitive hourly rate

Fully remote

Flexible engagement

High-impact work

Job summary

Jobgether seeks a Freelance GPU Kernels AI Task Auditor to audit GPU kernel tasks used to train and evaluate AI models from a fully remote location. You will assess realism, solvability, reproducibility, and testing criteria, and provide actionable technical feedback.

The role focuses on memory allocation issues, execution bottlenecks, and parallel computing logic errors, with emphasis on CUDA/Triton implementations and performance considerations. Remote work from international locations.

Qualifications

• Demonstrated experience in GPU kernel development and optimization.
• Expertise in CUDA or Triton programming, parallel computing, and hardware-level GPU performance optimization.
• Strong written communication skills with the ability to produce concise, actionable feedback.

Responsibilities

• Evaluate GPU kernel tasks for accuracy, realism, solvability, and reproducibility with reliable tests.
• Rigorous testing and troubleshooting of complex GPU scenarios to validate behavior.
• Audit tasks for memory allocation bugs, bottlenecks, synchronization issues, and parallel logic errors.
• Review CUDA, Triton and other GPU kernel implementations for correctness and practical applicability.
• Provide clear, specific feedback and recommended improvements.

Skills

GPU kernel development

CUDA

Triton

Parallel computing

Memory management

Performance optimization

Technical feedback

Tools

CUDA toolkit

Job description

Jobgether seeks a Freelance GPU Kernels AI Task Auditor to audit GPU kernel tasks used to train and evaluate AI models from a fully remote location. You will assess realism, solvability, reproducibility, and testing criteria, and provide actionable technical feedback.

The role focuses on memory allocation issues, execution bottlenecks, and parallel computing logic errors, with emphasis on CUDA/Triton implementations and performance considerations. Remote work from international locations.