Making research material easier to publish and check

The problem I was working through

At the University of Münster, the work joined two needs that often appear separately in research projects: a usable web application and source material that is structured enough for downstream analysis.

I worked as a Student Assistant from May to November 2025 in Münster. The project gave me a practical view of how delivery automation and data correctness can support research work without becoming the research itself.

Two tracks of work

The web track focused on a dynamic application using Python and Quarto, plus CI/CD workflows for the serverless frontend. That work connected the application with the process used to deliver it.

The data track focused on processing and validating 1,700+ pages of OCR output into conformant XML documents for downstream research analysis. That work made a useful distinction clear: a file can parse successfully and still need validation against project rules and source material.

Contribution overview

The diagram separates the data-processing work from the web-delivery work. It does not claim that the application directly consumed the XML output.

What the metric means

The 1,700+ page figure matters because OCR correction and XML validation are not only volume tasks. Each page also needs to keep a traceable relationship to source material, and each generated document has to satisfy the structural expectations of the project.

For me, the important lesson was the difference between three checks: well-formed XML, conformance to project rules, and fidelity to the original material. Passing one check does not automatically satisfy the others.

The engineering habit this reinforced

Research tooling benefits from explicit boundaries. A build can prove that a site renders. It does not automatically prove that source data was reprocessed, validated, or reviewed. A parser can prove that XML is well-formed. It does not prove the transcription is faithful.

That is why I like keeping the stages visible: data checks, rendering, CI/CD, and reused artifacts should each say what they prove and what they do not.

Sources and context

• Quarto publishing documentation for general publishing approaches.
• Python ElementTree documentation for XML parsing and serialization concepts.
• My OCR-to-XML note explains the distinction between parsing and validation.