Remote AI Evaluation Reviewer — Computer-Use Agent

AuraOne, Inc.; United States; Remote; USD 34,000 - 62,000; Full time

Job summary

AuraOne, Inc. is seeking remote Contractors to evaluate computer use agent outputs against our quality rubric. You will compare paired responses, label edge cases, and provide structured feedback to retrain models.

Clear written reasoning and reliable weekly availability are essential. Responsibilities include tagging unsafe content, calibrating against gold-standard examples, and documenting recurring failure modes for model improvement.

Qualifications

• Prior evaluation, annotation, or human-rater experience on computer use agent evaluation or adjacent content for Computer-Use Agent Evaluator work.
• Clear written reasoning that names the issue and the rubric clause being applied.
• Reliable async availability for at least 10 hours per week.

Responsibilities

• Evaluate computer use agent evaluation model outputs against a versioned rubric and assign severity tags for Computer-Use Agent Evaluator assignments.
• Compare paired responses and pick the stronger answer with a written rationale.
• Label hallucinations, instruction-following failures, and unsafe content with structured tags.
• Capture ambiguous prompts and route them back to the program team for rubric updates.
• Maintain reviewer-quality scores by calibrating against gold-standard examples each week.
• Document recurring failure modes so the modeling team can target them in the next training run.

Skills

Model output evaluation

Rubric-based annotation

Severity tagging

Inter-rater calibration

Computer Use Agent evaluation

Software engineering and computer use

AI evaluation

Rubric writing

Expert review

Job description

AuraOne, Inc. is seeking remote Contractors to evaluate computer use agent outputs against our quality rubric. You will compare paired responses, label edge cases, and provide structured feedback to retrain models.

Clear written reasoning and reliable weekly availability are essential. Responsibilities include tagging unsafe content, calibrating against gold-standard examples, and documenting recurring failure modes for model improvement.