AWS Trainium / NKI Kernel Expert

Anyone AI Inc.; Argentina; Remote; ARS 146,645,000 - 293,289,000; Part time

Job summary

Anyone AI Inc. is seeking experienced engineers to evaluate and improve NKI kernel implementations for AWS Trainium hardware. You will review, optimize, and benchmark kernel tasks, focusing on idiomatic Trainium usage and performance improvements rather than direct GPU translations.

The role is remote, part-time, project-based consulting. Candidates should have hands-on NKI kernel experience and familiarity with Trainium/Inferentia2 architectures and related tooling.

Qualifications

• 2+ years hands-on experience developing or optimizing NKI kernels.
• Experience with AWS Trainium and/or Inferentia2 hardware.
• Familiarity with Trainium architecture and GPU-to-NKI migration patterns.

Responsibilities

• Review NKI kernel correctness and Trainium-specific development patterns.
• Evaluate CUDA to NKI migrations and optimize performance.
• Benchmark memory management across SBUF, PSUM and HBM.
• Assess tile-based computation and DMA scheduling for efficiency.
• Provide technical feedback and quality assessment of kernel implementations.
• Ensure idiomatic and optimized Trainium code rather than direct GPU translations.

Skills

NKI kernel development

AWS Trainium / Inferentia2

Tile-based computation

DMA orchestration

Performance profiling

CUDA → NKI migrations

Numerical correctness across backends

Job description

Anyone AI is recruiting experienced AWS Trainium / Neuron Kernel Interface (NKI) engineers for a specialized project focused on evaluating and improving kernel development tasks for AI workloads.

We’re looking for engineers with hands‑on experience building or optimizing NKI kernels on AWS Trainium or Inferentia2 hardware who understand how Trainium’s architecture differs from traditional GPU programming.

What You’ll Work On

You’ll review and evaluate technical tasks involving:

• NKI kernel correctness and Trainium-specific development patterns

• CUDA → NKI kernel migrations

• Trainium performance optimization and benchmarking

• Memory management across SBUF, PSUM, and HBM

• Tile-based computation and DMA scheduling

• Cross-platform numerical correctness between CUDA/Triton and NKI

• Trainium-specific performance bottlenecks and optimization opportunities

• Technical feedback and quality assessment of kernel implementations

The work involves determining whether implementations are not only technically correct, but also idiomatic and optimized for Trainium hardware rather than simply translated from GPU-based approaches.

What We’re Looking For

• 2+ years of hands‑on experience developing or optimizing kernels with the Neuron Kernel Interface (NKI)

• Experience working with AWS Trainium and/or Inferentia2

• Strong understanding of:

  • Tile-based computation

  • SBUF / PSUM / HBM memory hierarchy

  • Partition dimension constraints

  • DMA orchestration

  • Trainium-specific optimization techniques

• Ability to evaluate CUDA → NKI migrations

• Experience profiling and optimizing workloads on Trainium

• Understanding of numerical differences across GPU and Trainium backends

• Strong ability to analyze complex technical implementations and provide clear written feedback

Nice to Have

• Experience with the AWS Neuron SDK or Neuron Compiler

• CUDA or Triton kernel development experience

• Knowledge of NeuronCore-v2 architecture

• Experience with FP32, BF16, FP8, and INT8 workloads

• Experience benchmarking workloads on Trn1 or Trn2 instances

• Familiarity with nki.language, @nki.jit, or XLA custom calls

• Experience with technical evaluation, AI/ML data projects, RLHF, or rubric-based assessment

Engagement

Work Type: Remote
Engagement: Part‑time, project‑based consulting
Focus: AWS Trainium / NKI kernel engineering and technical evaluation

This is a strong fit for engineers who have worked deeply with AWS Trainium infrastructure and low‑level ML kernel optimization and are interested in applying that expertise to technically challenging AI projects.