AI Copilot Prompt Engineer and Evaluator (Berkeley Research Group)
Developed and deployed custom AI Copilot agents for a state crypto licensing program, contributing prompt design and response evaluation for application document screening. Utilized Retrieval-Augmented Generation (RAG) as part of AI agent training to enhance document validation, eligibility checks, and workflow automation. Evaluated and refined text output quality to streamline state financial examiner workflows and improve applicant processing times. • Designed structured prompts and evaluated generated outputs for regulatory compliance applications. • Labeled and reviewed more than 100 documents per application for eligibility in the AI system. • Collaborated with stakeholders to define quality criteria for AI Copilot responses. • Participated in continuous improvement and data labeling for Copilot agent effectiveness.