Cultural institutions have invested heavily in linked data systems, and ontologies such as CIDOC-CRM are crucial for interconnecting information within cultural data space(s), allowing users to analyze multiple datasets horizontally and reconstruct the historical contexts of cultural objects. Yet the realization of this program depends not only on technological advancement but on creating an ecosystem accessible to the communities that produce and study the data. The distinction is not merely technical but epistemic: data may be structurally accessible while remaining interpretively opaque to those without the specialized knowledge to query it. As an alternative and complementary approach, this paper proposes human-centered mediation: a retrieval-augmented generation (RAG) system that allows users to explore CIDOC-CRM encoded heritage data through natural language, adapting the data to the logic of the user's inquiry rather than the reverse.