How to Make Museums More Interactive? Case Study of <scp>Artistic Chatbot</scp>

Conversational agents powered by Large Language Models (LLMs) are increasingly utilized in educational settings, in particular in individual closed digital environments, yet their potential adoption in the physical learning environments like cultural heritage sites, museums, and art galleries remains relatively unexplored. In this study, we present Artistic Chatbot, a voice-to-voice RAG-powered chat system to support informal learning and enhance visitor engagement during a live art exhibition celebrating the 15th anniversary of the Faculty of Media Art at the Warsaw Academy of Fine Arts, Poland. The question answering (QA) chatbot responded to free-form spoken questions in Polish using the context retrieved from a curated, domain-specific knowledge base consisting of 226 documents provided by the organizers, including faculty information, art magazines, books, and journals. We describe the key aspects of the system architecture and user interaction design, as well as discuss the practical challenges associated with deploying chatbots at public cultural sites. Our findings, based on interaction analysis, demonstrate that chatbots such as Artistic Chatbot effectively maintain responses grounded in exhibition content (60% of responses directly relevant), even when faced with unpredictable queries outside the target domain, showing their potential for increasing interactivity in public cultural sites.