Data Labeling / RLHF & Linguistic Evaluation (Project-based)
Performed RLHF-style response evaluation by ranking or assessing AI outputs using truthfulness, helpfulness, and safety criteria. Conducted sentiment and intent labeling for user queries and model responses to support preference modeling and quality control. Checked claims against reliable sources to verify accuracy for AI training and verification tasks. • Evaluated response truthfulness, helpfulness, and safety • Labeled sentiment and intent for queries and responses • Fact-checked claims using reliable sources • Supported high-accuracy evaluation for AI training datasets