An outstanding success: The SEMIC / BDVA workshop!

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! 

Semantic interoperability, which ensures data is exchanged with clear, shared meaning, is key to unlocking the value of data and breaking silos across Europe. Yet, aligning diverse data models remains a longstanding challenge. As new European data-related initiatives emerge (e.g. data spaces, SIMPL, AI Factories), the need for seamless interoperability is more urgent than ever. Meanwhile, there’s growing interest in using AI (especially LLMs) to support this, though real-world adoption and tested use cases are still limited. 

With this context in mind, SEMIC and BDVA reached out to their communities, inviting them to submit concrete use cases and real-world applications. The workshop was then shaped around the most relevant proposals received. This workshop aims to be just the starting point of a broader initiative to build a community around the topic, bringing together researchers, experts and industry to identify challenges, opportunities and next steps. 

Following the opening remarks and motivation by Daniel Alonso (BDVA) and Georges Lobo (EC / SEMIC), three keynote presentations set the stage for the workshop:  

• Ed Curry (University of Galway, Insight, BDVA vice-president) presented “The Symbiotic Relationship between Data Spaces and AI” and the need to consider an unified data and AI lifecycle to maximise the potential of GenAI.  

• Emilien Caudron (SEMIC) presented the different AI driven tools that SEMIC can offer, focused on a Data Modelling Chatbot incorporating RAG 

• Lorenzo Gabrielli (JRC) showed some prototypes and projects where JRC is working to combine RAG and LLM for interoperability and knowledge management 

A first round of presentations came from the research and academic community:  

• Arunav Das (King College London) presented how LLMs can be applied to support enterprise questions answering systems, in the field of telecom 

• Orfeas Menis Mastromichalakis (National Technical University of Athens) showed how LLMS can be applied to detect and contextualise harmful language and its application in the cultural heritage sector, an approach already adopted by Europeana 

• Rafiqul Haque (Insight – University of Galway) presented an LLM-Based Semantic Mediation Engine, an intelligent bridge to complement traditional approaches to connect different datasets and vocabularies. 

Industry perspectives followed:  

• Vladimir Alexiev (GraphWise) presented how to use LLM to generate in-depth column descriptors and its application in the Underpin project (manufacturing data space) 

• Suna Akbayır and Guven Fidan (Artech Consulting) explained how LLMs can be used to support the semantic interoperability needs of the Digital Product Passport 

• Tobias Jacobs (NEC Labs Europe) showed how LLMs can support semantic interoperability both for structured and unstructured data and its application to Smart Data Models and NGSI-LD.  

• Harri Ketamo (HeadAI) showed their work on using LLMs to automate the identification and tracking of emerging skill pockets in specific technological areas 

• Frédéric Bellaiche (Dawex) highlighted the importance of AI-Powered Semantic Hubs to address data conformity assessment across data spaces  

Finally, the public sector was represented by:  

• Ronald Kok (City of Rotterdam) explained how they are using LLMs to find relevant data for societal challenges, by adding semantics into catalogues.

Equally important as the expert insights was the engagement with the audience, whose perspectives enriched the discussion. This is why all speakers were asked to finish their interventions with two provocative statements, inviting the audience to express their agreement or disagreement and based on this, to run the discussion.  

There was a quite broad consensus that current tools are not enough (low level of satisfaction with semantic interoperability tools in current dataspaces: 2.1/5 for interoperability tools, 2.2/5 for discovery tools) and the need and suitability of AI tools for this purpose (94% agree that AI tools that facilitate reuse of standards and data models are critical to advancing interoperability and 96% believe that Generative AI helps bridge the gap between formal semantics and everyday language). Still, flaws inherent to LLMs must be considered (76% say hallucinations and concept drift will be major challenges).

All insights from the workshop will be compiled in a report to be published on the SEMIC and BDVA websites. The conversation will also continue in a dedicated follow-up session during the Data Week 2025 joint BDVA-SEMIC workshop. Stay tuned to the BDVA channels for more updates!

We also invite you to join the Interoperable Europe LinkedIn group to stay up to date and reach out to DIGIT-SEMIC-TEAM@ec.europa.eu.