Report following the “LLMs to Support Semantic Interoperability” workshop
A report was released following the great success of the joint workshop titled “Large Language Models to Support Semantic Interoperability”, which was held on 10 April 2025.
A report was released following the great success of the joint workshop titled “Large Language Models to Support Semantic Interoperability”, which was held on 10 April 2025, co-organised by the Big Data Value Association (BDVA) and the European Commission’s SEMIC initiative, part of DG DIGIT. The event brought together researchers, industry professionals, and public sector stakeholders to examine the growing role of Large Language Models (LLMs) in enhancing semantic interoperability across digital systems.
The workshop served as a forum for presenting cutting-edge techniques, tools, and real-world applications where LLMs are being used to align data models, enrich semantics, and facilitate interoperability across domains. Central to the event was a call for contributions from experts showcasing how LLMs can bridge gaps in data understanding, particularly through applications like intelligent metadata generation, automated data structuring, multilingual semantic mapping, and the integration of LLMs with knowledge graphs.
Participants engaged in lively discussion, exchanging insights and identifying emerging challenges. Key themes included:
- Pragmatic vs. perfect solutions: Several speakers advocated for “good enough” approaches—leveraging conversational AI, incremental mappings, and text-based heuristics—as more viable in the short term than striving for exhaustive standardisation.
- Governance integration: The need to merge AI lifecycle considerations with data governance frameworks, including provenance tracking and quality assurance, was a recurring point.
- Reliability and human oversight: Concerns were raised around LLM hallucinations, with a consensus emerging that human validation remains essential, especially in sensitive or high-stakes domains.
- Scalability and tooling: While promising prototypes exist—such as data modelling assistants and semantic summarisation tools—participants noted they are often still in the experimental phase and not yet broadly deployable.
- Sector-specific applications: Use cases in manufacturing, public-sector data exchange, and digital product passports illustrated the need for LLMs to incorporate domain expertise and hybrid methods.
Throughout the day, the importance of human-in-the-loop design, clear benchmarks, and machine-readable data formats was consistently emphasised. Attendees also highlighted the potential for synergy between LLMs and European data space initiatives, envisioning smart agents that facilitate automated mediation between systems.
With more than 160 participants, the workshop from 10 April co-organised by SEMIC and BDVA on “Large Language Models (LLMs) applied to Semantic Interoperability” was an absolute success! You can access the slide presentations below.
Next steps
The workshop concluded with a unanimous perspective, coagulating all points of view of speakers: LLMs as key enablers of next-generation semantic interoperability. However, they agreed that success depends not only on technical development, but also on clear governance, well-structured validation mechanisms and the development of best practices for an AI-ready Europe.
Future initiatives will focus on building a dedicated community to continue exploring this fast-evolving field, sharing lessons and co-developing solutions that balance innovation with reliability. As semantic interoperability remains a cornerstone of the European data strategy, the responsible integration of LLMs into this space promises to accelerate the development of cohesive, cross-border digital ecosystems.
On 28 May, from 14:30 – 15:30, Data Week 2025 will be hosting the “Facilitating semantic interoperability with LLMs” session, which is building on the workshop co-organised by SEMIC and BDVA. The session takes the discussion further into practical implementation. We’ll showcase ongoing projects and real-world applications of LLMs in addressing challenges such as semantic alignment, intelligent metadata generation, and cross-domain interoperability. The session will highlight how these technologies are being tested and adopted within data spaces to support trustworthy and interoperable digital ecosystems.