Managing Stress and Anxiety for Mental Health using Generative AI

Current Students: Keerthana Srinivasan, Sam Rahimzadeh Holagh, Lavanya Nidamanuri, Dhruv Srivastava, Aryan Sharma, Shikhar Sharma

Status: Current

The proposed research aims to design and develop a Virtual Reality Instructional (VRI) training environment using Unity, with interactive avatars powered by generative AI. This environment will focus on mental health support by enabling non-player characters (NPCs) to simulate supportive conversations, deliver therapeutic interventions, and guide users through mindfulness and cognitive behavioral exercises. This innovation can enhance the realism, personalization, and accessibility of mental health training and interventions. The objectives include:

To design and implement a VR-based training platform in Unity with AI-driven avatars.
To design and develop AI Therapist Companion and Stress-Relief Chatbot in AR.
Integrate Pose Estimation and Computer Vision Tasks for yoga exercises.
To integrate generative AI for natural language processing and dynamic dialogue systems.
To simulate realistic, supportive interactions that include therapeutic techniques like CBT and mindfulness.
To evaluate the effectiveness of the environment in promoting mental health awareness, reducing stress, and improving user engagement in training simulations.

Expected Outcomes:

A working VR prototype with AI-driven avatar interaction
Enhanced training for mental health professionals and users seeking stress/anxiety support
Evaluation data demonstrating efficacy and usability of the system
Potential for future development into clinical or educational applications

AI Assistant with ConvAI and VR	Conversational agents	UNT Health Center in VR
Chat Bot in Health Center	yoga exercises in VR	UNT Health Center in VR

Character Design and Personality Configuration

The virtual agent has been modeled after a licensed clinical psychologist with a background in Cognitive Behavioral Therapy (CBT). Attributes such as empathy, calmness, patience, and professionalism have been encoded to facilitate trust and open communication. All responses have been constructed using emotionally intelligent, reflective language. A slow and steady speech rhythm has been applied, and affirming phrases have been incorporated to convey understanding (e.g., “That’s understandable,” “You’re not alone in this”). Jane has been equipped with the Academic Anxiety Scale (Cassady, 2020), consisting of 11 items rated on a four-point scale. These items have been reformulated into conversational prompts and delivered sequentially. Scores are computed through simple summation, and anxiety levels are categorized as follows:

11–14: Not Anxious

15–20: Mild Academic Anxiety

21–29: Moderate Academic Anxiety

30–44: High Academic Anxiety

Publications

Summitt, A., Bhatt, B.J., Sharma, S.,"Emergency Assistive Mobile Application with Digital Twin for Real-Time 3D Navigation in Indoor Environments", Proceedings of the 23rd IEEE/ACIS International Conference on Software Engineering, Management and Applications (SERA 2025), Las Vegas, USA, May 29-31, 2025.