Culturally Aligned Generative AI to Support Para-Counselors in Low-Resource Community Mental Health Settings

This project aims to develop culturally grounded generative AI tools to support para-counselors in Bangladesh who provide front-line mental health support for anxiety, depression, and psychosis. Para-counselors are often embedded in community-based services but work with limited training and diagnostic tools, leading to oversimplified or incorrect attributions (e.g., financial hardship as the cause of all distress). Our goal is to integrate a causal reasoning framework with large language models (LLMs) to help para-counselors understand their clients using explanatory models that are culturally and contextually appropriate and align better with client experiences and scientific understanding, thereby informing the development of culturally resonant care plans and intervention strategies. We adopt a rigorous, participatory approach: lived experience experts are embedded throughout the project lifecycle-as co-researchers, evaluators, advisory board members, and compensated co-designers-to shape tool functionality, language, and deployment strategy. During the research, we will build on data centered on people with lived experiences, develop a causal graph library reflecting local idioms of distress, and evaluate language and reasoning performance with stakeholders. The solution will be open source, interoperable with existing systems, and designed for future scaling across LMIC contexts. Long-term capacity-building with partner organizations will ensure sustainability and continuous improvement.