Universidad Católica Luis Amigó: MetaExtract API in continuous production on DSpace and ABCD since 2024

Institutional context

Universidad Católica Luis Amigó is a private university based in Medellín, Colombia. Its information unit operates two different and complementary platforms: an institutional repository on DSpace for its open-access academic output, and a catalog on ABCD for managing the collection. It is a common configuration in the region, and also one of the most awkward to automate, because each system has its own data model and its own way of receiving information.

The challenge

The library needed to reduce transcription work in both systems, not just in one. A solution that only addressed the repository would have left the catalog as it was, and vice versa. That ruled out alternatives tied to a specific platform and called for a component that could be integrated into both workflows without forcing the university to unify its systems, something that was not on the table and would have been a project of a different scale.

There was also an implicit condition: the service had to last. An automation that works during implementation and degrades afterwards solves nothing in a library that works every day.

What we implemented

• MetaExtract API integrated into both environments, the DSpace repository and the ABCD catalog, with extraction tuned to the metadata schema of each.
• BiblioScan AI, added later for AI-assisted cataloging of physical material from photos.

How we did it

1. Assessment per platform. The fields, mandatory elements and document types of DSpace and of ABCD were surveyed separately, because the extraction output had to be different for each destination.
2. Model tuning. Extraction was trained with real documents from the institution, covering its own document types in addition to the usual academic ones.
3. Dual integration. The API was connected to both workflows, coordinating the handling of credentials and errors in each environment with the university's team.
4. Validation and production launch. Control batches were processed in both systems, and the human review procedure before approving each record was defined.
5. Expansion of scope. With the service already stabilized, BiblioScan AI was added for physical material, covering both the description of digital documents and the cataloging of the print collection.

Result

MetaExtract API has been running in production without interruption since October 2024, with the contract renewed and a second annual term under way. BiblioScan AI was added later and has also been renewed. That is the most relevant verifiable result: not a pilot that came to an end, but a service the library sustains in its daily operations across several budget cycles and on two platforms at the same time.

Technologies

MetaExtract API, BiblioScan AI, DSpace, ABCD, metadata extraction with language models, image recognition for cataloging, Dublin Core and MARC21.