ML Research Engineer: Inference & GPU Kernels

What you'll do

AfterQuery is building a research cohort in ML inference and GPU kernel engineering, covering inference serving systems, kernel optimization and large-model deployment infrastructure. The ad says plainly that this is research and evaluation work, not production engineering: you set the standard for "correct" and "excellent" on problems you know deeply. You design realistic technical scenarios and problem sets, author expert-level reference solutions and grading rubrics, and evaluate AI-generated outputs for correctness, depth and domain judgment.

Who fits

• Domain: machine learning and AI
• 6 to 10 years of experience
• A PhD or doctorate, or a master's with a strong research record
• Currently or recently hands-on in the field
• Strong written communication and comfort with independent, async, remote work
• Preferred: publications or recognized open-source work, experience reviewing others' technical work, and track-specific signals such as SGLang, vLLM, Mamba/Mamba2, TensorRT-LLM or GPU kernel work

Candidates are background-checked. The ad warns that those meeting fewer than 70% of the stated qualifications may be flagged for misrepresenting their experience, which could affect future project staffing.

What it pays

The board headlines $135/hr; the ad's "Why apply" text says $100 to $170/hr, based on experience. The listing does not publish its payment terms or weekly hours.

Worth knowing

Good:

• One of the higher bands on the board, for a narrow and in-demand specialty
• Remote and flexible; fits around existing work
• Evaluation work that draws on research-level depth

Less good:

• A demanding education and experience bar, with background checks
• Narrow specialty: general ML experience without inference or kernel depth is unlikely to match
• Payment terms and hours are not stated

About this listing

Read from the AfterQuery Experts board on 18 October 2026 (posted 4 August 2026). The board files it under software engineering. AfterQuery's help centre describes an unpaid, one-attempt assessment and ID verification for its expert pools. See AfterQuery.