Agentic Narrative Audio Guides: A Hybrid Multi-Agent System for Cultural Audio Guides

In this paper, we introduce Agentic Narrative Audio Guides (ANAG), a hybrid multi-agent system used to personalize narratives inside cultural Points of Interest: rooms, paintings, sculptures, monuments and everything that could be an attraction. ANAG grounds LLM generation with expert-curated descriptions and adapts tone and focus using a user profile based on its language, interest tags, and a self-description of the user. We present a coordinated multi-agent system that produces a structured narrative composed of a title, intro, story, and takeaways for immediate client playback. We present a proof-of-concept mobile application with real-world locations and discuss design choices. We discuss ethics, accessibility, and future possible extensions. Our contribution is a deployable, hybrid architecture that illustrates how LLM-enabled MAS can support attraction-level cultural engagement.