The article presents the results of a pilot project investigating the potential of automated manuscript analysis using large language models (LLMs) and HTR tools. The aim is to develop and evaluate a practical workflow, compare available solutions, and formulate good practices for future projects involving fifteenth-century handwritten sources. As a case study, a fifteenth-century formulary manuscript (BJ rkps 42), written in Gothic cursive by multiple scribes, has been selected. The manuscript poses significant challenges due to paleographical variability, numerous abbreviations, and linguistic complexity resulting from the interaction of Polish and Italian traditions.

The project consists of three modules. (1) Automated text acquisition (HTR): different infrastructures and engines are tested, along with various training strategies. The use of LLMs for post-editing and correcting transcription errors is also examined. (2) Translation and philological processing: several LLM are compared in terms of semantic accuracy, grammatical correctness, and sensitivity to administrative formulae, with particular attention to prompting strategies. (3) Information extraction: generation of regesta and extraction of structured data (persons, places, institutions), including TEI annotation and linking to external authority files.

The results make it possible to assess the real potential of automation and to propose a scalable workflow for processing fifteenth-century manuscripts.