shamiriAI – Scalable, Culturally-Informed AI for Fidelity Feedback and Adaptive Training in Youth Mental Health

Training and supervising non-specialists to deliver evidence-based mental health interventions (“task-shifting”) is a promising strategy for closing the youth mental health treatment gap. However, scaling this approach is constrained by supervision bottlenecks, as traditional methods are labor-intensive and rely on scarce professionals. To address this, we’ve developed shamiriAI, an AI-augmented supervision tool for non-specialist lay-providers. shamiriAI processes session audio, using multilingual Automatic Speech Recognition (ASR) and prosodic analysis to generate structured feedback for supervisors. Operating within an "AI-in-the-loop" design, shamiriAI supplements—rather than replaces—supervisors, who remain the primary agents of supervision. Our proposal seeks to further develop, deploy, and rigorously evaluate shamiriAI through three aims: (1) Further develop shamiriAI through co-design to ensure a safe, contextually-grounded system with advanced multilingual ASR and prosodic modeling; (2) Conduct a comprehensive implementation science evaluation by deploying shamiriAI within training and supervision workflows of a task-shifted youth mental health program; and (3) Conduct a well-powered randomized non-inferiority trial (N=1,000 youth; N=20 lay providers) to test if AI-augmented supervision maintains clinical effectiveness while improving fidelity and cost-effectiveness. Outputs include open-source models and datasets, implementation guides, and peer-reviewed manuscripts. Ultimately, we hope to create a template for safe, contextually-grounded AI-augmented supervision in mental health programs globally.