ELMTEX – Affordable, Privacy-Compliant AI for European Healthcare

The ELMTEX project delivers specialized AI solutions tailored to the European healthcare environment, with applications extending to security and e-government sectors. Our approach optimizes small-to-medium language models for domain-specific tasks, providing cost-effective alternatives to off-the-shelf commercial AI solutions.

Our Approach

ELMTEX emerged from a decade of research integrating machine learning into healthcare applications. We have developed techniques to process clinical documentation efficiently, particularly addressing the paper-based reality of many European healthcare institutions. Our three-pronged modeling approach (Naive Prompting, Retrieval-Augmented Learning, and LoRA Fine-Tuning) transforms unstructured medical text into standardized formats while maintaining privacy and security.

Our most significant finding shows that smaller, fine-tuned models often outperform larger models for specialized tasks, dramatically reducing hardware requirements and costs for healthcare providers. This makes AI implementation accessible even to institutions with limited IT budgets.

European Regulatory Compliance

The ELMTEX approach aligns perfectly with emerging European regulations. Our solutions support European Health Data Space (EHDS) requirements by facilitating data standardization and interoperability across systems. The on-premises deployment model ensures complete data sovereignty, addressing GDPR concerns while fulfilling EU AI Act requirements for transparency and explainability.

Multi-Sector Applications

While initially focused on healthcare, our technology offers significant benefits across sectors, for example:

Healthcare
Our solution extracts structured information from clinical documentation, reduces administrative burden, and integrates with existing health information systems – all while maintaining complete data privacy through on-premises deployment.

Security
Security agencies can process unstructured text sources while maintaining strict confidentiality requirements. The ability to deploy models entirely within secure environments ensures sensitive information never leaves controlled systems.

E-Government
Government institutions can automate document processing, extract relevant information from citizen submissions, and streamline administrative workflows while maintaining transparency and data protection.

Results and Future Direction

Our evaluations demonstrate that domain-specialized models can match or exceed larger general-purpose models at a fraction of the computational cost. Currently, we are conducting detailed assessments with clinical teams while exploring applications in manufacturing, financial services, and other sectors.

The ELMTEX project represents a European approach to AI innovation, balancing technological advancement with privacy protection, cost-effectiveness, and regulatory compliance to ensure the benefits of AI are accessible to all stakeholders in the European digital ecosystem.