Remote Data Labeling Specialist (Munich)

Eckdaten

München
ab 30 $
Training
vor 2 Tagen

Arbeitsmodell

Vollständig remote
Nur DE
Stellenbeschreibung

Remote Data Labeling Specialist (Munich)


Full-Time, Remote

Rex.zone is hiring Munich-aligned remote data labeling specialists to produce high-quality training data and evaluation signals for AI systems. You will label and evaluate text, image, and multimodal datasets used in LLM training pipelines, RLHF workflows, prompt evaluation, and content safety labeling with structured QA evaluation and feedback loops.

What You Will Do

  • Perform data labeling across NLP and computer vision annotation tasks (classification, ranking, multimodal labeling)
  • Create RLHF-style preference rankings and provide clear rationales aligned to rubrics
  • Run prompt evaluation for helpfulness/harmlessness and other quality dimensions
  • Complete named entity recognition labeling for domain-specific datasets
  • Execute content safety labeling using policy categories and careful judgment
  • Participate in QA evaluation cycles: sampling reviews, gold-set checks, disagreement resolution, and calibration sessions
  • Flag ambiguous guidelines, document edge cases, and incorporate updates to improve consistency

Required Qualifications

  • Experience with data labeling and/or QA evaluation in a metrics-driven workflow
  • Strong attention to detail and high annotation guidelines compliance
  • Clear writing and reasoning skills for ranking and evaluation tasks
  • Comfort working asynchronously in a distributed remote team

Preferred Qualifications

  • Familiarity with RLHF, LLM evaluation programs, or evaluation dataset creation
  • Exposure to content safety labeling policies and calibration processes
  • Experience with computer vision annotation tools or NER workflows

How To Apply

Apply on Rex.zone to complete a short screening, a guideline-based assessment, and a QA evaluation calibration step. Selected candidates will be assigned to projects supporting LLM training pipelines.