The LLMs4EU initiative (“Large Language Models for the European Union”) is a 3-year European project (2025-2028) funded under the Digital Europe Programme and coordinated by ALT-EDIC. Its aim is to strengthen Europe’s independence, competitiveness, and cultural diversity in generative AI by building a transparent, multilingual ecosystem of Large Language Models (LLMs). 

LLMs4EU is a multi-national, multi-institutional initiative aimed at developing trustworthy, multilingual large language models across different EU sectors. Notilyze is a core industrial partner with operational experience in deploying AI systems at scale, making them uniquely positioned to guide technical and domain-specific decisions. 

Problem description – challenge 

Companies who want to bring Large Language Models (LLMs) (but also AI in general) into production, see the following three challenges: 

• Accuracy: the LLMs do not meet the company’s standards in terms of performance; whenever asking questions to AI which represent daily operations within the company, the LLM returns too generic answers or hallucinates confidently.
  
  
• Specific knowledge: the LLMs in the market often lack knowledge on specific domains and/or languages, making them not suitable for a tangible business use case. This has multiple reasons, but most importantly 1) training data on the specific use case is not available and 2) companies offering LLMs are not interested in finetuning a model to a very specific niche with a small TAR
  
  
• Regulatory alignment: implementation of AI also involves implementing new compliance measures, and certain applications of LLMs require a level of transparency in the models that cannot be given by the models that are currently in the market, as this is the IP of such a model. Compare it to Coca Cola sharing its recipe for their soft drink; the value is in this secret, not in the ingredients necessary to make the soft drink.
  
  A combination of these factors prevents companies from getting value out of AI quickly; return on investments should be quick to convince board members about the value AI can bring to their organization. 

Solution 

Notilyze collaborates in the Large Language Models for the European Union (LLMs4EU) project to solve all of the above, specifically in the context of implementation within the EU. The project’s main objectives are:  

• Data infrastructure / “one-stop shop” for language data: Collecting and processing language data across all official EU languages (and regional/low-resource ones) to support the training of large language models (LLMs)
  
  
• Tailored models & tools for sectors/languages: Enabling development of language models that are adapted for specific languages, domains, sectors and use cases — lowering the barriers to entry for smaller actors. 
  
  
• Sectoral use-cases: in line with the previous bullet, concrete deployments are done within the project in five defined sectors: energy, telecom, tourism, public services, science. 
  
  
• Evaluation & benchmarking infrastructure (“Virtual Evaluation Centre”): The project plans a platform for transparent evaluation/benchmarking of LLMs in the European context. 
  
  
• Open access / tools for SMEs: Ensuring that European companies (especially SMEs) can access tools and resources for training/adapting models, rather than only large incumbents. 

The resulting LLM products will be fully open: 

• Open model weights
• Open code
• Open training data

This approach helps companies to obtain models that are fully transparent. Also, the data used for training must be compliant to EU regulations (think of GDPR, AI Act, the Dutch AVG). This helps companies to implement AI within their organizations that are EU compliant by design.  

How? 

The project consists of different Work Packages (WPs), which are sequentially executed by the partners involved. These WPs cover the whole process of bringing a LLM into production: 

• Define a use case with validation metrics.
• Choose a base model and define the appropriate finetuning template.
• Obtain data relevant for training and start training.
• Evaluate the model on validation metrics, biases and compliance to EU regulations.
• Deploy the model in the context of the use case.
• Offer model in the open access repository for European organizations and citizens to be used.
  
  

Notilyze’s role

Notilyze has a role in finetuning work package. They investigate different finetuning methods and how to evaluate them. This is currently done specifically in the Public Services domain, of which Notilyze is the use case domain coordinator. In that role Notilyze makes sure that use cases which are in scope, are addressed properly; the solution should add direct value to the end user.  

Notilyze already has experience in finetuning LLMs for a specific use case; auditing annual reports of banks to check for CSRD compliance.  

Outcome 

Currently the focus is on the following use cases: 

• Public Administration assistant to help employees within governments and/or citizens with their questions on documents published by the government:
  
  
• A chatbot to help citizens with information requests on a level of local municipalities either directly or by helping them navigate within a government e-portal;
  
  
• a tool that asks victims of crimes a set of questions in order to direct them to the correct ‘counter’, based on their answers, so they can be helped quickly by the prosecutor.
  
  
• A tool to help researchers and practioners get updated about latest Health Care developments available in public (or private literature / databases) with Natural Language queries, evidence-base responses & smart reporting features. 
  
  
• A LLM-driven offline transcription of Irish language audiovisual content for a Public Broadcasting service

Current use cases have been suggested in different languages: GR, IT, PL, RO, PT, FR, LU, LV and EN. 

About Notilyze Cloud
Notilyze Cloud is a specialized hosting provider for SAS® Viya®, offering secure, scalable, and compliant environments tailored to organizations that rely on advanced analytics. As home to the Notilyze SAS Viya Centre of Excellence, the company combines deep platform expertise with operational excellence to deliver high-performance, future-ready cloud solutions for SAS users worldwide while meeting the highest standards of data integrity and regulatory compliance.

For more information, please contact:

📧 info@notilyze.com
🌐 www.notilyze.com