How Much AI Can Research Data Management Handle?

How Much AI Can Research Data Management Handle?

Working with research and measurement data places especially high demands on data quality and reliability. That is exactly what makes the question so interesting: How much AI can data management in these fields actually handle?

That was the topic of our discussion yesterday at the meeting of the Technical Working Group (AK Technik) of Measurement Valley e.V., which this time was hosted at our offices.

We shared insights into our current work on two concrete questions: How can LLMs help to unlock unstructured data (e.g. from PDFs) and integrate it meaningfully into a data management system? And, as a next step, how can LLMs make searching for data easier?

For us, it is essential that AI is not an end in itself. The use of LLMs, for example, has to be critically examined. Does the technology really add value at this particular point? How can data quality and traceability be ensured? And how can risks relating to digital sovereignty and data security be addressed?

Our approach is this: we use LLMs in data management for specialized tasks where they can deliver real added value. In doing so, we look closely at where small and local models can be used sensibly — particularly from the perspectives of digital sovereignty and cost efficiency. We also deliberately build in ways for humans to review and control LLM queries and results.

The exchange in the AK Technik once again showed how valuable it is to discuss these questions from different perspectives. Many thanks to all colleagues at Measurement Valley for the inspiring discussion.