This short paper discusses the evolution of a dataset in the pursuit of creating FAIRand accessible data (Wilkinson et al. 2016). The dataset concerns the DigitalSEE (Digital South-Eastern Europe) project, an initiative aimed at building a comprehensive digital repository of sources related to the Balkans during the Ottoman Empire, from the 16th to the 19th century. The project works with travelogues, epigraphic inscriptions, woodcuts, etchings, and other historical materials. The source languages include Latin, Greek, French, Italian, German, English, and Ottoman Turkish. Authors such as Felix Kanitz, Captain Schad, Mary Montagu, and others are featured (Kanitz 1875; Schad ca. 1721). The project team consists of historians, archaeologists, classicists, digital humanities specialists, and developers.

One of the objectives of the project is source tracing and prosopographic research. For example, when an epigraphic inscription is mentioned in a travelogue, the team investigates whether additional information about it exists, such as its repository if it is an artifact, or its remains if it is an archaeological site. One case studied in the project is that of Trajan’s Arch, where new findings have shed light that challenges the established narrative (Valchev 2024). Another major aim of the project is to create a data repository. At the beginning of the project, custom, non-proprietary, and well-documented XML was used to capture the full complexity of the documentation, resulting in more than 45 tags defining each file in the collection. These tags ranged from the different names a place or object may have, to the period the author believed it belonged to, to georeferencing coordinates and references to authoritative gazetteers such as GeoNames and Pleiades for places from antiquity. Researchers also selected one of nine predefined categories for each site or artifact, along with relevant subcategories, for instance, inscriptions may be funerary, boundary, list, votive, legal, milestone, etc., while settlements may be categorized as mahalla, çiftlik, tell, villa, village, town, and so on.

These predefined categories help ensure consistency across all files in the dataset. If a traveler mentions a votive inscription in a village, the researcher can assign two categories and subcategories to remain faithful to the source. Bibliographic references are also provided to document the origin of the information. The sourceContent element contains a portion of the original text entered by the researcher and it is included to support later use in topic modeling, advanced NLP tasks, and close-reading analyses. Each record also contains comments and remarks reflecting the researcher’s expertise regarding the source. When available, links to IIIF manifests (International Image Interoperability Framework) and VIAF (Virtual International Authority File) identifiers are provided as reliable authority references. Images of objects are also included and visualized.

For file management, JSON master files were used for ease of use. Researchers do not work directly with the raw XML, but with a Flask-based web application used for data entry, editing, preliminary geographic visualization, and statistical insights helpful to the research team. It also serves as a quality-assurance tool: certain fields are required, formats must be followed, and incomplete data cannot be submitted. Once validated, the data are sent to the public frontend of the application, built on the AIAX system (AIAX n.d.), which serves as a data-visualization and content-management platform.

The dataset was already findable, accessible, and semi-interoperable thanks to its comprehensive documentation. To make it fully FAIR, the team recognized the absence of established data standards for this type of historiography. This issue was addressed by creating a fully interoperable and reusable XML data structure compliant with the TEI P5 Guidelines (TEI Consortium 2023). To do so, we aligned our structure with the schemas and recommendations of the EpiDoc initiative, which provides a standard for encoding epigraphic and historical textual information in TEI (Elliott et al. 2006–2022), and which best matched the existing dataset without requiring excessive custom solutions.

Quality control was achieved through two main methods. First, all output files underwent automated schema validation against a custom TEI P5 RNG schema, ensuring compliance with the TEI EpiDoc guidelines and confirming structural and syntactic validity. Second, after automated transformation, the files were manually inspected and corrected for anomalies that automated checks might have missed. Version control and traceability were ensured by storing the final TEI-compliant files in a GitHub repository, documenting and tracking all changes and corrections (Chen et al. 2025; Ram 2013).

The collection, although small, contains 150 rich files. These were transformed automatically using a Python algorithm built with xml.etree.ElementTree (Lundh 1999–2008). For careful file handling and versioning, the algorithm retrieves and formats the current date using a dedicated function. The workflow reads an XML file, processes it, and outputs the extracted data in TEI-compliant format.

A separate Streamlit application was developed to help researchers manage and search through the TEI-compliant XML files. It offers a simple search interface, options to view XML code, download individual entries, and access comprehensive metadata. The parsed plain text is displayed for readability by non-technical users, and metadata such as author, publisher, and date are extracted. The tool also functions as a quality-assurance mechanism, ensuring that TEI XML files maintain a uniform and consistent structure in their presentation of commentaries and editions.

The FAIRification of the DigitalSEE dataset ultimately reflects a broader commitment to thoughtful engagement with both historical material and emerging technologies. By transforming a small, multilingual, and highly heterogeneous corpus into a fully interoperable and reusable resource, the project foregrounds the importance of careful annotation, transparency, and interpretive responsibility in the creation of digital heritage collections. This process not only enhances access and sustainability but also invites new forms of interaction: between researchers, communities, and computational tools, while remaining attentive to questions of context, authorship, and the evolving role of AI in mediating cultural knowledge.