Design, architecture and safety evaluation of an AI chatbot for an educational approach to health promotion in chronic medical conditions

This paper presents the design, architecture, and safety evaluation of an AI chatbot tailored for educational purposes in man aging chronic medical conditions, focusing on Type 2 Diabetes Mellitus (T2DM). Leveraging conversational agents in health literacy, the chatbot integrates medically informed information, constrained responses, and response traceability to ensure appropriateness and compliance with protocols. By utilizing ChatGPT with retrieval augmented generation (RAG) and careful prompt engineering, the system ensures reliable, traceable, and privacy-conscious interactions. Safety and efficacy testing revealed just one inappropriate response (5%) in a simulated patient conversation and 15 (75%) fully appropriate responses. This study highlights the potential of AI chatbots in enhancing patient autonomy, reliability, and privacy in accessing medical knowledge for chronic conditions.