Content Moderation / Toxicity Detection
Specific tasks included text classification, multi-label annotation, and content moderation labeling, where text data was categorized into classes such as toxicity, hate speech, harassment, and safe content. Additional responsibilities included image annotation (bounding boxes and classification), data validation, and quality review of annotated datasets. The project involved handling high-volume datasets (10,000+ data points) across multiple domains, requiring consistency, speed, and attention to detail. Strict quality assurance measures were followed, including adherence to detailed annotation guidelines, maintaining high inter-annotator agreement, participating in review cycles, and consistently achieving >95% annotation accuracy, ensuring reliability and usability of the labeled data for downstream model training and evaluation.