Designing Human-Centered Generative AI Conversational Agents For Digital Cultural Heritage: A Case Study On Europeana

Digital cultural heritage platforms such as Europeana provide access to extensive and diverse collections but continue to face challenges related to searchability, knowledge discovery, and user experience. This thesis explores how generative AI- powered conversational agents can mediate access to such collections, addressing limitations of keyword-based search, metadata heterogeneity, and interface complexity. The study adopts a human-centred, mixed-methods case study approach. A prototype LLM-based conversational search agent was designed and developed using tool-augmented function calling to interface directly with Europeana’s public APIs. Methods included user research, prototype development, and a user study combining task-based testing, questionnaires, think-aloud protocols, and system log analysis. Empirical analysis focused on user interactions, perceived relevance, understanding, discovery, trust, control, and satisfaction, alongside technical inspection of agent behaviour. Particular attention was paid to how the agent handled missing or inconsistent metadata and how LLM planning and tool use shaped outcomes. Results indicate that conversational interaction can enhance exploratory search and discovery, lower cognitive barriers, and support vague or complex queries. However, significant limitations emerged due to the prototype’s system architecture and metadata incompleteness and inconsistency. The findings suggest that while generative conversational agents can meaningfully augment cultural heritage search, their effectiveness is tightly constrained by metadata quality and responsible system design. Future progress requires advances in both AI agent architectures and cultural heritage data infrastructures.