Digital tools offer novel opportunities to support healthcare skill acquisition, particularly through serious games designed to train caregivers. In this article, we present a hybrid narrative engine architecture that combines symbolic rules, tree structures, and the generative capabilities of large language models (LLMs). This approach addresses the need to create interactive scenarios that focus on the specific challenges of care management. At the core of the system, rules and story trees ensure narrative coherence, while LLMs generate contextualized text, enhancing interaction with more credible non-player characters (NPCs). The engine is designed to provide a customisable user experience, with real-time feedback and scenario unfolding that adapts to the player’s choices and progress. Future developments include extending the approach to other mental health disorders, improving authoring tools, and establishing a clinical evaluation framework to accurately assess the therapeutic impact.