How European Governments Are Using Open AI Models to Build Sovereign Public Services
Governments need modern digital solutions that improve public services while keeping sensitive information under control. Open AI models offer a way forward by giving public-sector organizations more control over model evaluation, local-language adaptation, infrastructure, safeguards and data governance.
Some open models also make their training datasets publicly available. This allows governments and developers to assess whether the data was authorized, identify potential bias and evaluate compliance requirements.
NVIDIA partners and European governments are working together to deploy advanced AI models that improve government efficiency while maintaining control over data and infrastructure. These efforts include localized datasets, open models, agent systems and tools for evaluating AI applications in public-sector environments.
The NVIDIA Nemotron-Persona Collection includes partner-based datasets used by NeMo UserSim, NVIDIA’s open-source library for population-scale user simulation. The library helps organizations dynamically evaluate and train models and AI agents in local languages.
European Commission Expands Managed AI Applications
Berlin-based software company Deepset provides the Haystack Enterprise Platform, the foundation of the European Commission’s AI@EC platform. The platform enables organizations to build and run AI applications on commission-managed infrastructure.
The update brings together agent development, execution, optimization and governance. It also adds capabilities for comparing operational costs, evaluating answer quality and tracking issues across AI applications.
NVIDIA Triton Inference Server supports shared services used to find and rank relevant information. This builds on operational requirements demonstrated in deployments such as AI@EC, where reusable components, role-based access, audit trails and guardrails support document summarization, knowledge management and multilingual policy analysis.
Poland Uses Advanced Computing for Healthcare and Employment Services
Deloitte is helping build an agent system for health and employment services in Poland using an open model.
Poland’s eHealth Center is building an intelligent medical-imaging service platform with NVIDIA DGX BasePOD and the MONAI framework. The platform helps doctors analyze X-rays, CT scans and MRI images, detect abnormalities and prioritize urgent cases, reducing the time patients wait for results.
With support from Deloitte, Poland’s Ministry of Family, Labor and Social Policy is developing an agent factory on NVIDIA accelerated computing infrastructure. Designed to support more than 400 labor bureaus, it runs models such as PLLuM, GPT OSS and Inkling for document search, summarization and digital assistants. These tools help staff assist job seekers more efficiently.
NVIDIA is also working with SpeakLeash in Poland and the University of Ljubljana in Slovenia on country-specific additions to the synthetic persona program, with localized datasets to follow.
UK Dataset Supports Regional AI Model Evaluation
London-based Lokai Lab released Nemotron-Persona-UK, a dataset containing 1 million synthetic personas that reflects the United Kingdom’s demographic and geographic diversity while helping mitigate privacy risks.
Potential applications include medical assessments and bias testing across age groups, education levels and rural and urban populations. For developers building public-service models, the dataset provides a way to test whether models work consistently across different communities.
Germany Develops Digital Assistants for Sensitive Workloads
In Germany, Materna is introducing Sinfonie, an on-premises platform that combines generative AI, voice interactions and 3D avatars. NVIDIA Audio2Face is available for animation.
Built on a German state government project, the service allows employees to access organizational knowledge through verbal questions while storing data on the organization’s own infrastructure.
For organizations handling sensitive information, secunet, NVIDIA, Red Hat and Deepset are developing sovereign cloud products built on secunet’s SINA Cloud foundation. The platform includes a security component authorized to use sensitive information.
Planned products include NVIDIA accelerated computing and the NVIDIA Nemotron open model, Red Hat OpenShift AI for deploying and managing AI workloads, and Deepset Haystack for AI-agent orchestration. The infrastructure will be operated by secunet in a German data center.
To support regional development, the Sovereign Open Source Foundation Models Research Initiative, or Soofi, develops a family of AI models for Germany’s industrial and public sectors.
Nemotron-Persona-Germany is an open dataset containing 1 million records, each with six synthetic personas based on the 2022 Census across all 16 federal states and 400 districts. It was created with NVIDIA NeMo Data Designer.
Because the dataset does not use personal data from real people, developers can use it to test whether public-service models work consistently across regions, age groups and communities. Like the UK version, it can support model evaluation and testing. NeMo UserSim allows developers to evaluate and train models with simulated users based on synthetic German personas.
France Brings Open AI Models to Public-Sector Workspaces
Selected by CANUT, France’s public purchasing agency for digital services, Linagora and Docaposte offer a sovereign collaborative workspace built on Twake.ai.
More than 2,500 eligible organizations, including local authorities, schools and healthcare organizations, will be able to access the framework without running their own tenders. The workspace provides civil servants with email, document sharing and collaboration tools. NVIDIA technology supports audio transcription, while the infrastructure is hosted in France.
The Occitanie Region is a preferred partner for expanding the deployment.
Greece Builds a Localized AI-Agent Ecosystem
Greek AI and energy company Kiefer is developing an agent factory built on NVIDIA DGX SuperPOD with DGX GB200. The project uses the Sophea open-weight model and Greek-language tools for knowledge retrieval, voice assistants and digital assistants.
These capabilities help Greek-speaking public-sector teams find and use agency information while giving organizations greater control over where agents run and how data is processed.
NVIDIA Open Models Codefest Helps Teams Build Public-Sector AI
To help public-sector organizations integrate agentic AI in Europe and beyond, the NVIDIA Open Models Codefest brings together startups, software providers and research labs across the region. The initiative focuses on turning open models into locally controlled solutions for challenges such as medical inference, AI guidance and model governance.
The top five teams are Pleias, zenla eye, Opposition, Algorithmica GmbH and ABA technology.
Deutsche Telekom and T-Systems provided the computing infrastructure for teams to train and deploy language models at scale.
Source: blogs.nvidia.com