This keynote traces a research and commercialization journey that began in large-scale deception detection and automated interviewing systems and culminated in RhetorixLab, an AI-resilient assessment platform designed to restore authenticity and trust in an era of generative AI. Drawing on years of work in video-based feature extraction and behavioral signal processing, the talk highlights how rich multimodal data from voice, language, and facial expression can be captured and analyzed using commodity hardware and cloud-scale infrastructure. The discussion then pivots to higher education, where traditional text-based assessments are rapidly losing diagnostic value, and shows how asynchronous oral assessments reintroduce human judgment, communication skill, and genuine reasoning.
