AIWorkJobs
Mercor logoMercorPlatform listing

Trainium (NKI) Kernel Expert

Review Neuron Kernel Interface (NKI) development work that supports training and evaluating frontier AI models. You will judge kernel correctness, CUDA-to-NKI migration fidelity, Trainium performance choices, and…

Remote · United States onlyHourly
Work arrangement
Remote
Eligibility
United States only
Engagement
Hourly
Compensation
$70–$90/hr
Specialty
AI Training
Weekly commitment
40 hours per week

About the role

Review Neuron Kernel Interface (NKI) development work that supports training and evaluating frontier AI models. You will judge kernel correctness, CUDA-to-NKI migration fidelity, Trainium performance choices, and numerical consistency across GPU and Trainium hardware, then write clear feedback against defined rubrics. The work calls for hands-on NKI experience with AWS Trainium or Inferentia2.

What you would do

  • Evaluate NKI development tasks for quality, correctness, and fit with the target hardware.
  • Assess how faithfully developers migrate CUDA kernels to NKI.
  • Review performance choices for AWS Trainium, including NeuronCore pipeline use, tensor-engine throughput, and memory-bandwidth limits.
  • Assess numerical correctness across GPU and Trainium, including accumulation order, rounding, and mixed-precision behavior.
  • Write clear, rubric-based feedback on the work.

Requirements

  • At least 2 years of hands-on kernel development or optimization with NKI on AWS Trainium or Inferentia2.
  • Strong knowledge of NKI patterns, including tile-based computation, SBUF, PSUM, and HBM memory management, partition-dimension constraints, and DMA orchestration.
  • Experience evaluating the quality of CUDA-to-NKI migrations.
  • Familiarity with Trainium performance profiling, including NeuronCore pipeline utilization, tensor-engine throughput, and memory-bandwidth bottlenecks.
  • Experience defining or evaluating numerical-correctness standards across GPU and Trainium, including accumulation order, rounding behavior, and mixed-precision semantics.

Preferred qualifications

The source marks these as preferred, not required.

  • Direct experience with the AWS Neuron SDK, Neuron Compiler internals, or NKI kernel library contributions.
  • Prior CUDA or Triton kernel development.
  • Knowledge of Trainium hardware, including NeuronCore-v2 architecture, on-chip SRAM topology, and FP32, BF16, FP8, and INT8 data types.
  • Experience benchmarking ML training workloads on Trn1 or Trn2 instances.

Who should apply

Kernel engineers with at least 2 years of NKI experience on Trainium or Inferentia2 may be a strong fit, especially those who can evaluate CUDA migrations, profile Trainium performance, and assess cross-platform numerical correctness.

How applying works on Mercor

Create a profile, complete the role checklist and AI interview, then verify your identity before a project offer.

  1. Create an account and confirm your résumé.
  2. Complete the role checklist, written assessment, and AI interview where the listing requires them.
  3. Submit the application and wait for review or matching.
  4. Complete a work trial and identity verification before onboarding.

Typical timeline: The AI interview is about 20 minutes; work trials are often 4–6 hours; identity checks are usually minutes.

Paid steps: The work trial is paid.

Assessment and interview requirements vary by role. Check the checklist inside the specific listing.

Process checked Nov 19, 2026. AIWorkJobs does not submit applications or charge a fee. Read the full Mercor process and its official source on the Mercor page.

Skills mentioned by the source

  • neuron kernel interface
  • aws trainium
  • aws inferentia2
  • cuda
  • triton
  • aws neuron sdk
  • neuron compiler
  • kernel development
  • kernel optimization
  • tile-based computation
  • sbuf
  • psum
  • hbm
  • dma
  • neuroncore-v2
  • numerical correctness

Listing reference: trainium-nki-kernel-expert-dLDIZ6. Source: Mercor. If something here contradicts the source, the source controls. Report a correction.

Related roles

Other work on the index.

See more in this category

Mercor

GPU Kernel Expert

Review GPU and accelerator kernel-development tasks that support training and evaluation of a frontier AI lab’s models. Judge numerical correctness, benchmark fairness, task scope, and compilation and runtime validity across varied kernel work. You’ll also write rubric-based feedback that explains your findings.

Location and eligibility
Remote · United States only
Engagement type
Hourly
Compensation
$70–$90/hr
  • cuda
  • triton
  • nki
Machine Learning & AIView job

AfterQuery

ML Research Engineer — Inference & GPU Kernels

AfterQuery is assembling senior experts in ML inference and GPU kernel engineering to assess how frontier AI models handle difficult, real-world technical problems. You’ll define standards for correct, high-quality work through research and evaluation, not production engineering. The role spans inference serving, kernel optimization, and large-model deployment infrastructure, and calls for 6–10 years of experience plus a PhD or a master’s degree with a strong research record.

Location and eligibility
Remote · Eligibility not specified
Engagement type
Not specified
Compensation
$135/hr
  • machine learning
  • ml inference
  • gpu kernels
Machine Learning & AIView job

Meridial by Invisible

Technical Task Auditor

Audit technical tasks used to train and evaluate AI systems as an independent contractor. You will focus on one specialty that matches your experience, such as GPU kernels, Kubernetes, machine learning, or application security. Your reviews will check task quality, reproducibility, tests, and evaluation criteria, and you will provide actionable feedback on any problems.

Location and eligibility
Remote · Eligibility not specified
Engagement type
Not specified
Compensation
$60/hr
  • gpu kernels
  • aws trainium
  • nki
Quality Assurance & TestingView job

micro1

Member of Technical Staff, Model Training (infra & platform)

Own the systems that let researchers train models, run large experiments, and deploy specialized AI systems. The work spans distributed training, GPU infrastructure, inference, simulation, and agent execution, with architectural decisions that affect research speed and model performance. This is a full-time remote role requiring deep distributed-systems experience and hands-on work with GPU-based machine learning infrastructure.

Location and eligibility
Remote · Eligibility not specified
Engagement type
Full-time
Compensation
$7–$8/hr
  • Distributed ML Training Infrastructure
  • GPU Performance & Inference Optimization
  • Distributed Systems & Platform Architecture
Machine Learning & AIView job