This poster presents a controlled workflow for AI-supported historical research developed in the context of the Repertorium Academicum (REPAC) project at the Historical Institute, University of Bern. Using nodegoat as a research environment, the workflow connects structured prosopographical data on European academics between 1250 and 1550 with AI-supported methods for information retrieval, data enrichment, and analysis. Two complementary workflows are highlighted. The first focuses on source processing and data enrichment, including translation of historical texts, named entity recognition, text tagging, and structured extraction into nodegoat. The second combines curated REPAC data with large language models through a retrieval-augmented generation pipeline: relevant data are retrieved first and then supplied as contextual grounding for dialogic analysis. The poster argues that AI can support discovery in historical research when embedded in transparent, data-driven environments. At the same time, it reflects critically on possible effects of accelerated AI-assisted workflows, including reduced depth of subject engagement and fewer serendipitous discoveries. The central principle is: retrieve first, interpret second.
Kaspar Gubler (2026) studied this question.