“HIVEMIND is a highly ambitious and technically very demanding project.”
— Georg Rehm, DFKI
In this interview, Georg Rehm from the Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) shares how his team contributes to HIVEMIND’s most data-intensive tasks. As leaders of several use cases and responsible for data-driven workflows, DFKI brings extensive experience in large language models and data technologies. Georg also reflects on the project’s potential to support digital sovereignty and reshape software development processes across Europe.
Watch the interview
Prefer reading?
You can find a full transcript of the interview below.
Hi, this is Georg from the German Research Centre for Artificial Intelligence, DFKI, here in Berlin.
We’re participating in HIVEMIND with our research group on Speech and Language Technology. In the project, we’re responsible for a number of different things, most importantly the work package on the data-driven workflow, providing data gathered from the use case partners to the actual system.
We’re also responsible for use case planning and coordination within HIVEMIND.
How does your expertise contribute to the objectives of HIVEMIND?
In the past, we’ve coordinated or participated in a number of EU and national projects that are highly relevant to the HIVEMIND approach, including work on data curation, data technologies, and large language models: developing them from scratch, evaluating them, and applying them in use cases.
What excites you most about being part of the project?
The most exciting part about being in this project, and about the project itself, is that there are so many aspects I could mention.
First of all, it’s a great interdisciplinary team collaborating toward a common goal. We have a very diverse and ambitious set of use cases. The project as a whole is highly ambitious, technically very demanding, and extremely interesting.
We’re training language models not only on general data but also on data provided by the use case partners, working toward a system that helps support the software development life cycle through agentic AI. That is very, very exciting.
What impact do you expect HIVEMIND to have on industry and society?
In terms of impact, we’re aiming to develop a lightweight and adaptable solution that can support our partners, and eventually, external clients and organisations working in software engineering, by simplifying and improving their processes.
The idea is that different LLMs trained on domain-specific data can interact with each other across the development life cycle to co-create and assist at different stages. This concept of agentic AI could really enhance and streamline how software is built.
If we manage to pull this off, to develop something lightweight, customisable, and easy to deploy on-premises for clients, the impact could be immense. It would be particularly meaningful in terms of promoting Europe’s digital sovereignty, which is more important than ever in today’s geopolitical context.
At DFKI, we’re also heavily involved in various data space initiatives, especially the Language Data Space. One of our goals is to use this infrastructure to collect and manage data from use case partners and potentially even from clients. This would give us a GDPR-compliant infrastructure for data collection and training of LLM agents, enabling a solid foundation for HIVEMIND’s architecture. It’s a novel and exciting use case for data spaces, and we’re really looking forward to seeing it take shape.