Legal-Domain AI Evaluation & Training (Source verification, evaluation rubrics, audit/prompt frameworks)
Provided source-verification procedures to assess legal citations and detect unverifiable information in AI outputs using a strict hierarchy of statute, precedent, case law, and doctrine. Designed evaluation rubrics and scoring criteria to standardize the assessment of legal-output quality and accuracy. Built audit and prompt-design frameworks to support multi-step review for AI-assisted legal analysis. • Source hierarchy enforcement and unverifiable-citation flagging • Rubric creation for legal quality and accuracy scoring • Multi-step audit/review workflow design for AI prompts • Anti-fabrication behavior aligned with hallucination-control concepts