Remote Preference Dataset QA Evaluator RLHF

AuraOne; United States; On-site; USD 55,104 - 82,656; Full time

Job summary

AuraOne is seeking a remote Preference Dataset QA Reward Model Evaluator to review prompts and model outputs against our quality rubric. You will compare paired responses, label edge cases, and write structured feedback to help retrain the model.

The role focuses on identifying hallucinations and instruction-following failures, tagging with the correct policy categories, and calibrating judgments against gold-standard examples.

Qualifications

• Prior evaluation, annotation, or human-rater experience on preference dataset QA reward model evaluation or adjacent content.
• Comfort applying multi-page rubrics consistently across long batches.
• Clear written reasoning that names the issue and the rubric clause being applied.
• Strong attention to detail and the ability to flag when a prompt itself is the problem.
• Reliable async availability for at least 10 hours per week.

Responsibilities

• Evaluate preference dataset QA reward model evaluation model outputs against a versioned rubric and assign severity tags for 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

Preference Dataset QA Reward Model

Preference ranking

RLHF

Rater calibration

Preference

Dataset

Job description

AuraOne is seeking a remote Preference Dataset QA Reward Model Evaluator to review prompts and model outputs against our quality rubric. You will compare paired responses, label edge cases, and write structured feedback to help retrain the model.

The role focuses on identifying hallucinations and instruction-following failures, tagging with the correct policy categories, and calibrating judgments against gold-standard examples.