Rexzone · India
About The Role
You will support AI/ML teams by executing and improving data labeling and evaluation workflows used in LLM training pipelines across NLP, named entity recognition, computer vision annotation, and content safety labeling. You will follow annotation guidelines compliance, document edge cases, and help reduce ambiguity so models learn consistently.
What You Will Do
Perform data labeling for text and/or vision datasets to meet training data quality targets
Complete RLHF-style preference ranking and response quality scoring to support model performance improvement
Run prompt evaluation and QA evaluation to identify errors, hallucinations, policy violations, and formatting issues
Apply named entity recognition and taxonomy tagging using clear annotation guidelines compliance
Conduct content safety labeling for sensitive, policy, and restricted content categories
Audit datasets for consistency, duplicates, label noise, and distribution gaps; propose fixes
Write concise rationales and escalate ambiguous cases to improve guidelines
Track throughput and quality metrics (accuracy, agreement rates) and participate in calibration sessions
Collaborate with engineers and QA to validate changes to labeling rubrics and evaluation checklists
Support experimentation by comparing model outputs across variants and documenting findings
Required Skills
STEM fundamentals and strong analytical reasoning
Comfort with Python and SQL for basic data checks (filters, counts, joins) and simple scripts
Understanding of NLP and/or computer vision annotation concepts
Experience applying annotation guidelines compliance and documenting decisions
Ability to perform QA evaluation, spot inconsistencies, and propose improvements
Clear written communication for rationales and issue reporting
Working Model
Remote, distributed work with QA calibration and clear workflows.
High accuracy and consistency are essential to improve training data quality.
