Demo: Can Visual Stimulation Enhance Reminiscence-Therapy Chatbot?

Sofiia Kononovych, Ostap Kilbasovych, Kateryna Bilyk, Yuliia Vistak, Zhiwen Fan, Junyuan Hong

Keywords: Reminiscence Therapy, Chatbots, Large Language Models, 3D Memory Visualization, Emotional Engagement, Human–AI Interaction, Mental Health

TL;DR: We present Remi-Visual, a chatbot that combines LLMs with 3D memory visualization to enhance reminiscence therapy and boost emotional engagement.

Abstract:

Reminiscence Therapy (RT) supports emotional well-being by encouraging individuals to recall meaningful past experiences, often guided by trained facilitators and visual stimuli. However, access to RT is limited, and traditional approaches lack scalability. To address these challenges, we developed \textbf{Remi-Visual}, a chatbot that combines Large Language Models with real-time 3D memory visualization. Unlike prior RT systems that rely on pre-existing images, Remi-Visual dynamically collects autobiographical details during conversation and generates personalized visual reconstructions using DALL·E 3 and ViewCrafter.
In this work, we conducted a between-subjects study, comparing two chatbot versions: with memory visualization (Remi-Visual) and without (Remi-Non-Visual). Results show that Remi-Visual significantly increased participants’ feelings of interest and showed strong positive trends for enthusiasm and excitement, willingness to reuse the chatbot. Open-ended feedback highlighted the engaging nature of visuals, while pointing to the need for improved alignment with personal memories.
These findings suggest that integrating 3D visual reconstruction into conversational agents can enhance user engagement and emotional outcomes in RT.
Our work demonstrates the potential of AI-driven multimodal tools to expand access to therapy and support emotional resilience.

Submission Number: 126