Beyond the Chatbot: Building Safe, Emotionally Intelligent AI for Mental Health Support with Longitudinal Memory

Session Overview:

Generative AI can hold a conversation, but healthcare requires much more than fluent responses. Mental-health support demands context, memory, risk recognition, clinical safety and clear escalation when human intervention is needed.

In this case study, Dr Imraan Jhetam will share the development of Sakinah, an AI emotional-support companion powered by the CALMOS cognitive architecture designed to bring together emotion, context, memory, risk, safety and reasoning before our own LLM CALMOS AQL generates its response.

The session will explore how clinical thinking has been translated into an AI system designed to listen, understand and remember while maintaining safety and governance around every interaction.

Using practical examples from the development and deployment of Sakinah, the session will examine how AI can provide continuous emotional support between traditional healthcare encounters without attempting to replace clinicians.

The session will cover:

• Moving beyond single-session chatbots towards longitudinal memory and continuity of support.
• Detecting changes in emotional state, distress and risk across conversations.
• Building safeguarding, escalation and clinical governance into the AI architecture.
• Separating conversational fluency from clinical safety and decision-making.
• Designing multilingual and culturally appropriate emotional-support services.
• The role AI can play alongside NHS mental-health services, clinicians and existing care pathways.

The case study will also discuss the practical lessons involved in moving from an AI concept to a deployable healthcare service, including safety, explainability, clinical oversight and the boundaries that healthcare AI must respect.