Culturally Aligned Generative AI for Para-Counsellors
Culturally Aligned Generative AI to Support Para-Counsellors in Low-Resource Mental Health Settings
Syed Ishtiaque Ahmed · University of Toronto · Canada
What they're working onOur project develops culturally aligned generative AI to support para-counsellors delivering front-line mental health care for anxiety, depression, and psychosis in low-resource settings in Bangladesh. Generic LLMs, trained largely on Western data, misread local idioms of distress (such as "tension") and impose explanatory models that don't fit clients' realities. We integrate a hierarchical causal knowledge graph — built from lived-experience narratives and expert annotation — with LLMs through a GraphRAG architecture, so the assistant can reason about culturally grounded cause–effect pathways and give para-counsellors contextually appropriate, causally grounded guidance. People with lived experience are embedded throughout as co-designers, causal-model validators, usability testers, and members of the Lived Experience and Ethics Advisory Group. Over 24 months we will build a validated Bangladeshi knowledge base, prototype and benchmark the AI, co-design the interface, and run a six-month field pilot, alongside a co-produced Responsible AI and ethical governance framework. All causal models and fine-tuning methods will be released open source for adaptation to other Global South contexts.
AI approach
LLMs · causal knowledge graphs · GraphRAG · culturally aligned GenAI
Mental health focus
Anxiety · depression · psychosis · frontline care
Links