Harbor Expert Network
Multimodal QA Reviewer
$80–$115/hrpay
Required skills
About Harbor
Harbor connects domain experts to the development of frontier AI models. Real-world expertise is turned into training data, evaluations, and feedback loops that improve how models perform. AI labs and enterprises use Harbor to train models and build reliable agents through advanced evaluations and field capture programs. Experts contribute directly to how AI systems learn, reason, and perform across robotics, computer vision, wearable AI, industrial inspection, healthcare, and agentic systems. Harbor identifies and vets top talent through an AI recruiter, enabling high-quality contributions at scale.
Job description
Job title: Multimodal QA Reviewer
Job type: Contractor
Location: Remote
Job summary: Validate multimodal labels across video, audio, and text for production vision and understanding models. Your domain judgment catches systematic drift before it reaches training.
Key responsibilities
- Audit multimodal annotations for temporal alignment, class correctness, and guideline fit.
- Adjudicate edge cases involving occlusion, rare classes, and ambiguous scene context.
- Measure and improve inter-annotator agreement through targeted review samples.
- Flag pre-label systematic errors and recommend active-learning priorities.
- Support taxonomy updates with concrete examples from real review queues.
Required skills and qualifications
- Senior computer vision or video understanding background (research, MLOps, or QC lead).
- Experience with detection, segmentation, or video classification QA at scale.
- Strong grasp of class imbalance, hard negatives, and long-tail failure modes.
- Ability to work asynchronously with written QA verdicts and severity tags.
Preferred qualifications
- Experience with audio-visual sync review or multimodal foundation-model evals.
- Familiarity with COCO-style taxonomies and custom enterprise ontologies.