More guidelines, more complexity
In this project, we set out to make European cardiology guidelines more accessible, interactive, and easy to maintain — while preserving full digital sovereignty.
Staying current with cardiology guidelines is vital for effective patient care. However, the sheer volume and complexity of these guidelines often overwhelm clinicians, making it challenging to keep up.
A) Comparison of total recommendation items in current and previous versions of esc guidelines on valvular heart disease (VHD), adult congenital heart disease (GUCH / ACHD) and infectious endocarditis (IE). B) While previous document have mostly employed plain text and tables, current guidelines more often rely on images, flowcharts, and infographics.
The problem with consumer chatbots
While consumer focused large language models can support understanding, they come with important limitations like outdated knowledge, unclear and unctrollable sources, hallucinations and inherent biases—issues that hinder reliable clinical decision-making.
Our system combines the reasoning of large language models with guideline data, offering answers and multimodal context with full control over the knowledge base.
Our System bridges the gap by combining the intelligence of LLMs with verified ESC guideline data. With multimodal retrieval for text, images, and audio, it delivers precise, context-aware insights. Clinicians retain full control over updates, ensuring the information stays accurate and relevant.
How it works
Data Processing: guideline and recommendation texts (converted to machine readable format) and multimedia inputs (PDFs, audio, images) are structured into searchable chunks.
Storage: Data is securely stored in advanced vector and document databases.
Hybrid Retrieval: A semantic search system fetches relevant context for seamless clinical queries. It doesn’t even matter what language the documents are in because the system searches for logical connections and not words.
Reasoning and answer generation: State of the art large language model are used (interchangeable) for reasoning and text generation while the source content is displayed in separate tabs for review.
Full control: Admins can instantly update the knowledge by adding new guidelines the minute they are published and remove others once obsolete. This ensures the system stays up-to-date and relevant. Admins maintain full control over the content, enabling precise customization for clinical needs.
Choose your model. Keep your knowledge. Stay in control.
This approach allows hospitals, societies, and research networks to deploy AI-powered access to any guidelines while retaining full control over content, provenance, and compliance with European data protection frameworks.
Our System demonstrates how European institutions can leverage generative AI responsibly — combining accessibility, transparency, and sovereignty in clinical decision support.