A Real-Time Voice Agent for Delivering Individual Cognitive Stimulation Therapy to Persons Living with Dementia
Individual Cognitive Stimulation Therapy (iCST) is a home-based psychosocial intervention for persons living with dementia (PLwDs) whose therapeutic value depends not only on activity content but also on how the session is conducted. Existing digital and voice-based iCST systems can still feel staged, especially around pauses, restarts, and slow endings. We describe a browser-based research prototype that uses continuous, interruptible speech to deliver a short iCST session. A central design move in the prototype is to separate local turn-taking from session-level control: the underlying real-time API manages interruption and exchange timing, while the application manages overall session progression. We describe how a prior formative study, iCST guidance, dementia communication literature, and iterative implementation informed the session arc, the system boundary, and the role-and-prompt structure. We conclude by discussing the design implications, limitations, and future evaluation needs for conducting an iCST session with a real-time voice agent.