The main goal of this chapter is to provide tools that facilitate the automatic transcription of content from passport registry books in the Portuguese National Archives. This will enable the extraction of relevant information from the transcribed text to populate the CIDOC-CRM Knowledge Base, ultimately allowing public access to search and explore its content using natural language queries. This chapter proposes an approach for developing an automatic transcription system to process passport requisitions. This approach leverages fine-tuning, a widely adopted technique in machine learning, within the open-source deep learning toolkit PyLaia. The Arkindex platform was used for the preparation of task-specific datasets. Since manually annotating manuscript text is both expensive and time-consuming, assessing system performance is crucial to determine the required size of the manually annotated corpus for achieving adequate accuracy. This chapter also presents a study to establish the appropriate size of the training dataset, specifically the number of annotated images of passport requisitions needed.