Senior Kernel Engineer: 26-02794

Akraya, Inc.; Bellevue (WA); On-site; USD 117,000 - 124,000; Full time

Job summary

Akraya, Inc. is seeking an experienced kernel optimization engineer to develop and optimize low-level compute kernels for AI accelerators in Bellevue, WA.

You will ensure training environments reflect real-world kernel engineering challenges by writing, debugging, and tuning operators, while using profiling tools and accelerator interfaces to validate task realism. The role requires hands-on work with CUDA/Triton or NKI and strong coding skills, with a background in high-performance compute and

Qualifications

• 7+ years of relevant experience required.
• Experience with AWS Trainium/Inferentia and NKI.
• Hands-on proficiency with CUDA, Triton, or NKI.
• Familiarity with Neuron SDK, Neuron Profiler, and framework integration (PyTorch, JAX).
• Experience porting kernels between accelerator backends (e.g., CUDA/Triton to NKI).

Responsibilities

• Develop and optimize custom compute kernels for AI accelerators.
• Validate training environment tasks for realistic kernel development scenarios.
• Debug and profile kernel code to identify and resolve performance issues.
• Analyze kernel performance against hardware constraints and memory hierarchy.
• Consult with training environment teams on task realism and quality.

Skills

CUDA programming

Triton optimization

Profiling tools

Debugging kernels

Performance tuning

Education

Bachelor's degree in CS/CE/EE or equivalent

Tools

AWS Trainium/Inferentia

Neuron Kernel Interface (NKI)

Neuron SDK

Neuron Profiler

PyTorch

JAX

Job description

Primary Skills:

• CUDA programming (advanced)
• Triton optimization (advanced)
• Profiling tools (advanced)
• Debugging kernels (advanced)
• Performance tuning (advanced)

Contract Type: W2

Duration: 3+ months with possible extension or conversion

Location: Bellevue, WA (#LI-Onsite)

Pay Range: $85.00 - $90.00 Per hour on W2 #LP

Job Summary

This role involves the low-level development and optimization of compute kernels for AI accelerator hardware. You will ensure training environments accurately mirror real-world kernel engineering challenges by writing, debugging, and optimizing operators. A key aspect of this position is using profiling tools and accelerator programming interfaces to validate task realism and define solution quality.

Key Responsibilities

• Develop and optimize custom compute kernels for AI accelerators.
• Validate training environment tasks for realistic kernel development scenarios.
• Debug and profile kernel code to identify and resolve performance issues.
• Analyze kernel performance against hardware constraints and memory hierarchy.
• Consult with training environment development teams on task realism and quality.

Must-Have Skills

• Computer Science, Computer Engineering, Electrical Engineering, or equivalent technical degree and 7+ years of relevant experience is required.
• Direct experience with AWS Trainium/Inferentia and the Neuron Kernel Interface (NKI) and Hands-on proficiency with a low-level kernel programming language such as CUDA, Triton, or the AWS Neuron Kernel Interface (NKI).
• Familiarity with the Neuron SDK, Neuron Profiler, and framework integration (PyTorch, JAX).
• Experience porting kernels between accelerator backends (e.g., CUDA/Triton to NKI) and understanding of deep learning operator implementations (attention, GEMM, normalization).

Industry Experience

Experience in developing and optimizing low-level code for AI accelerators is essential, with a strong preference for experience with Triton over CUDA.