Technical Task Auditor

What you'll do

Someone else writes a technical task meant to train or test an AI system; you check whether it holds up. The listing names five things to assess: is the task accurate, realistic, solvable, reproducible, and backed by reliable tests and evaluation criteria. Where it falls short, you write clear, actionable feedback so the author can fix it.

The skill in this review work is spotting what is wrong with a task quickly: a test that passes for the wrong reason, a setup that only works on the author's machine, a "correct" answer that is not the only correct answer.

The eight specialties

You are matched to one track based on your experience, and you do not need all eight. The listing's tracks are:

• GPU Kernel Task Auditor: GPU kernels
• TRN Kernel Task Auditor: AWS Trainium and NKI
• CVE Vulnerability Task Auditor: CVE and application security
• ML Challenge Auditor: machine learning
• AI Developer Trace Auditor: AI-assisted developer workflows
• AWS Serverless IaC Task Auditor: AWS Serverless and Infrastructure as Code
• Kubernetes Task Auditor: Kubernetes
• SWE-Bench Task Auditor: SWE-Bench

The same eight specialty areas appear as separate Mercor listings, such as its Kubernetes Task Auditor and SWE-Bench Task Auditor, at $70 to $90 an hour but restricted to US residents. Neither platform names its client, so we cannot say whether it is the same project; for engineers outside the US, this listing is the one that states no country requirement.

Who qualifies

• Hands-on experience in at least one of the eight specialties
• Passing a technical assessment in that specialty, which the listing says is part of selection
• The ability to explain a problem in writing so someone else can act on it

The listing gives no years-of-experience bar, degree requirement or country requirement.

What it pays

$60 an hour, the highest rate on the Invisible listings we read and well above the $17 of the general evaluation work. It is also above the $35 paid for writing terminal and infrastructure benchmark tasks, which is the authoring side of this kind of work.

The contract terms say payment is weekly via Stripe or Wise, and that Invisible cannot support H-1B or STEM OPT contractors.

Worth knowing

Good:

• A strong fixed rate for review work
• You only need depth in one specialty
• Reviewing is usually faster per task than writing, and uses judgment you already have
• Remote, on your own schedule
• No country requirement stated, unlike Mercor's US-only equivalents

Less good:

• A technical assessment on top of Meridial's resume upload (about 5 minutes) and unpaid qualification step (about 30 minutes)
• The listing does not describe the assessment, its length or whether it is paid
• No minimum amount of work, and no indication of volume per track
• Independent contractor terms: no benefits, your own tax to manage

About this listing

Read on the signed-in Meridial marketplace on 26 September 2026, posted about two weeks before we read it. The specialties, assessment requirement, rate and contract terms are the listing's own; the comparisons with other tracks and listings are ours. Invisible's terminal and infrastructure task-writing listing is the authoring counterpart, and our guide to AI training work for software engineers compares the wider field. See Invisible Technologies.