Trustworthy LLM-based Conversation Agents to Enhance Migrant Youth Mental Health

Project Leads

Reza Samavi, Roberto Sassi, Venkat Bhat

Team Members

Hirad Daneshvar, Alyssa Li, Vaishali Meyappan, Saeideh Mousavi, Shuting Xie

Over the last decade, the demand for youth and family mental health services has dramatically increased. This issue is further compounded for migrant youth and children, who face unique mental health challenges while navigating the complexities of integrating into a different society like Canada.

The main objective of this research is to investigate the trustworthy and ethical considerations of using conversational agents in reducing depression and anxiety symptoms in youth migrants and develop methods to provide specific guarantees of privacy protection, robustness and confidence of such agents.

1. What are the challenges for conversational agents interacting with immigrant youth who do not have the language skills or who carry different cultural backgrounds?
2. How can privacy-preserving and robustness mechanisms be integrated into conversational agents to safeguard sensitive mental health data while adapting to the diverse cultural, linguistic, and socio-economic contexts of migrant youth, minimizing biases and ensuring equitable outcomes?
3. What ethical framework mechanisms are necessary to improve reliability and foster trust in conversational agents among migrant youth and their caregivers while ensuring fairness, accountability, and inclusivity in their deployment?

This research follows a structured methodology in four distinct phases and employs a mixed methods approach combining computational experiments, user-centred design, participatory method and iterative evaluation.

The project is in progress with Phase 1, User-Centered and Participatory Design, currently ongoing.

Expected completion date: March 2027

Publications
Rueda, A., Hassan, M., Perivolaris, A., Teferra, B., Samavi, R., Rambhatla, S., Wu, Y., Zhang, Y., Cao, B., Sharma, D., Krishnan, S., & Bhat, V. (2025). Understanding LLM scientific reasoning through promptings and model's explanation on the answers. arXiv.

Teferra, B., Perivolaris, A., Hsiang, W., Sidharta, C., Rueda, A., Parkington, K., Wu, Y., Soni, A., Samavi, R., Jetly, R., Zhang, Y., Cao, B., Rambhatla, S., Krishnan, S., & Bhat, V. (2025). Leveraging large language models for automated depression screening. PLOS Digital Health.

Daneshvar, H., & Samavi, R. (2025). GNN’s uncertainty quantification using self-distillation [Preprint]. In Proceedings of Artificial Intelligence in Healthcare (AIiH 2025).

Rueda, A., Perivolaris, A., Roy, N., Weston, D., Shaya, S., Cote, Z., Ivanov, M., Teferra, B., Wu, Y., Rambhatla, S., Sharma, D., Greenshaw, A., Jetly, R., Zhang, Y., Cao, B., Samavi, R., Krishnan, S., & Bhat, V. (2025). Estimating quality in therapeutic conversations: A multi-dimensional natural language processing framework. arXiv.

Parkington, K., Teferra, B., Rouleau-Tang, M., Perivolaris, A., Rueda, A., Dubrowski, A., Kapralos, B., Samavi, R., Greenshaw, A., Zhang, Y., Cao, B., Wu, Y., Rambhatla, S., Krishnan, S., & Bhat, V. (2025). Human vs. LLM-based thematic analysis for digital mental health research: Proof-of-concept comparative study. arXiv.

Events and presentations:

• "GNN’s uncertainty quantification using self-distillation," presented by H. Daneshvar and R. Samavi, at Artificial Intelligence in Healthcare (AIiH 2025), Cambridge, United Kingdom, September 8–10, 2025.

Youth mental health; migrant children; LLM conversational agents; depression and anxiety