AI4RDCs – AI for Research Data Centers

Research data centers (RDCs) protect sensitive microdata, such as survey or administrative data. For this reason, internet access there is often deliberately restricted, and connections to public artificial intelligence (AI) providers are neither possible nor permitted.

As a result, researchers and RDC staff work without AI support, even though it could be especially helpful for complex metadata, access rules, extensive documentation, and programming. The AI4RDCs project aims to close this gap with a sovereign, open AI assistant.

The assistant is designed specifically for the questions and content requirements of research data, for example by drawing on local, curated sources such as metadata catalogs, documentation, guidelines, and code repositories. Among other things, it helps with:

• Understanding data: quickly grasp the structure, content, and limitations of microdata
• Finding suitable data: identify appropriate variables and subpopulations and check whether the research question and the data fit together
• Documenting analyses: explain and summarize scripts, make analysis steps traceable
• Writing and improving code: create and optimize analysis code
• Finding rules and procedures: answer questions about an RDC’s access rules and procedures in an understandable way

The project is aimed at the following target groups:

• Researchers who work with sensitive data and want to reach their results faster
• RDC operators and staff who want to support their users better and free up capacity
• Other infrastructures and research groups facing the same data protection requirements

The project is being developed at the LMU-ifo Economics and Business Data Center, the research data center at the ifo Institute – Leibniz Institute for Economic Research at the University of Munich. The center already has extensive experience in providing reusable infrastructure for secure research environments through infrastructure as code. The project is currently in its initial phase, evaluating which infrastructure and software components, including AI models, are best suited to the typical use cases. During the project, which runs until the end of 2028, the specialized capabilities for research data management will then be developed.