Owning the AI trust and ethics research program as the sole researcher, from survey analysis that surfaced motivation collapse and AI trust as primary user barriers to a human-centered framework that guides agent governance and product strategy

Translating research into production AI behavior by authoring the instruction and knowledge-base architecture for CCP’s chat-based intake and coaching agents, encoding escalation, disclosure, and error-correction rules through human-in-the-loop review

Owning release readiness for CCP’s AI intake agent, live on Instagram and moving to the company website, with a launch bar of under 5% claim error and near zero on prices, discounts, and program terms

Designed a staged evaluation pipeline (synthetic-persona pilot, live participants, retest) across three sprints in 13 days; the pilot predicted a pricing-disclosure defect that live participants confirmed within one week, and surfaced language-accessibility needs for non-native English speakers

Built a hallucination tracker coding 834 agent claims against the live site (AI-assisted, human-verified), traced errors past a knowledge-base contradiction to an instruction layer overriding it, and cut pricing errors from 31% to 0% (0 of 87 claims) across eight releases

Built a Claude-based evaluation setup guide that a teammate replicated on his own device to run agentic sessions independently, facilitated AI learning sessions for internal teams, and built an external intermediate AI tools class for CCP group sessions, backed by a Mural learning library