NVIDIA English-Domain RLHF: Acted as the QA of QAs; auditing primary reviewer outputs and adjudicating disputed evaluations to ensure high-consistency for critical training data.
Autonomous UI Review (OpenAI): Performed QA on a novel 3D modeling pipeline where AI operated Blender and GIMP. Verified step-by-step actions against screen recordings.
High-Stakes Evaluation: Conduct second and third-pass reviews on LLM reasoning and safety alignment, validating logic and instruction adherence.
Pipeline Improvement: Flagged systemic annotation errors, improving inter-reviewer reliability and audit accuracy.