NKI Kernel Task Auditor – AWS Trainium / Inferentia

Location: Remote – India

Employment Type: Contract

Compensation: USD $25–$40/hour

Commitment: 5–6 hours per day

Duration: Expected average of approximately 3–6 months; exact duration is not guaranteed

About the Role

We are looking for an experienced NKI Kernel Engineer / Kernel Task Auditor to support a specialized AI evaluation project involving AWS Trainium and Inferentia hardware .

In this role, you will review and audit AI-generated kernel engineering tasks, including native NKI implementations and CUDA-to-NKI migrations. You will assess whether implementations are technically correct, reproducible, performant, and appropriate for the target accelerator hardware.

This is a hands-on technical role requiring actual experience developing and optimizing kernels using the AWS Neuron SDK and Neuron Kernel Interface (NKI) .

What You’ll Do

- Review NKI kernel implementations and CUDA-to-NKI migrations.
- Identify logic errors, numerical issues, and hardware-inappropriate implementation choices.
- Validate test coverage and benchmark methodology.
- Investigate memory, execution, and performance bottlenecks.
- Assess whether kernel tasks are realistic, solvable, and reproducible on the intended accelerator.
- Review implementation quality against technical evaluation rubrics.
- Provide clear written technical feedback.




- Evaluate kernel optimization decisions including memory movement, tiling, DMA scheduling, and execution behavior.
- Review performance measurements and profiler results.

Required Experience

- You should have personally written and optimized NKI kernels using AWS Trainium or Inferentia2 hardware.

Hands-on experience should include:

- AWS Trainium – Trn1 and/or Trn2
- AWS Inferentia2
- AWS Neuron SDK
- Neuron Kernel Interface – NKI
- nki.language
- @nki.jit
- NeuronCore execution
- Trainium performance optimization
- Tile-based kernel programming
- SBUF
- PSUM
- HBM
- Data movement and DMA scheduling
- NKI profiling
- Kernel benchmarking
- CUDA-to-NKI migration

Evidence of Work Is Required Candidates must be able to provide public technical evidence demonstrating their own NKI work, such as:

- GitHub repositories
- Exact kernel files or commits
- Open-source contributions
- Kernel benchmarks
- Technical publications

You should be able to clearly identify:

- What you personally implemented
- Target hardware used
- Kernel or benchmark involved
- Performance results achieved

Solid Advantages Experience with any of the following would be particularly valuable:

- AWS Annapurna Labs
- Contributions to aws-neuron
- Publicly available NKI kernels
- Trainium performance publications
- Upstream NKI contributions
- Significant kernel optimization or benchmarking work

📌 NKI Kernel Task Auditor – AWS Trainium / Inferentia (India)
🏢 OttoMate
📍 India