A talk from one of our project partners - thymia
Voice-Based Psychiatric Assessment in the Real World and the Case for Production-Aligned Speech Frontends
Speakers
Stefano Goria, PhD
Chief Technology Officer & Co-Founder, thymia
Stefano leads thymia's engineering and ML research teams, building voice AI and speech biomarkers for mental and physical health. PhD in Theoretical Physics (University of Turin), over eight years as a quant at Citi and J.P. Morgan (heading structured products quant research, EMEA), shipping production ML systems including autonomous market-making agents. Publications span particle physics, reinforcement learning for biodiversity conservation, speech biomarkers, and mathematical foundations of machine learning. Builds robots in his spare time to experiment with embodied intelligence.
Agnes Norbury, PhD
Principal Research Scientist - thymia
Agnes has a PhD in cognitive neuroscience and held research fellowships in Computational Psychiatry at the University of Cambridge, Mount Sinai Hospital, and UCL before moving into applied digital mental health. At thymia, she has driven work on psychological validity and data quality for mental wellness models, baking clinical expertise into our depression and anxiety model, and scaling training datasets from tens of thousands to 75k+ unique speakers. She has also led an AI Quality taskforce and contributed to thymia's Safety Evaluation Framework for conversational agents in health and wellbeing. Her favourite work sits at the intersection of psychological science and machine learning: bridging insights from both to create robust, fair and useful digital health tools.
George Fairs
Machine Learning Researcher - thymia
George has a background in both engineering and cognitive science and is passionate about using his skills in shaping the future of health technologies. At thymia, he has helped design, train, and evaluate the company's core biomarker models, along the way contributing to clinically publishable research. He works within the full model lifecycle - from experimental design and rigorous validation through to production deployment - ensuring that models meet the high bar required for clinical application. Working closely with science and engineering, he helps bridge research insights and real-world clinical impact, translating cutting-edge methods into robust tools. His core interest lies in pushing the boundaries of what voice and language biomarkers can achieve for health assessment.
