AI output evaluation and data annotation for LLM training (RLHF)
Evaluated AI outputs and performed data annotation to support LLM training with RLHF-style workflows. This work involved applying quality rubrics and explicitly checking for hallucinations in the generated responses. The goal was to create higher-quality, more reliable training data for LLM behavior. • Evaluated 200+ prompts • Created and applied quality/rubric criteria • Detected hallucinations in model outputs • Produced labeled judgments for downstream training