AI-supported case processing at the Danish Health Authority

The Radiation Protection Unit at the Danish Health Authority has many responsibilities. One of them is supervising radiation sources. This concerns both patient safety, e.g. for dentists and chiropractors, and public safety, ensuring that people can wait for the bus in the same spot every day even if there’s a radiation source just behind the wall. And, of course, it’s also about safeguarding the staff who handle them - researchers, healthcare staff, and emergency personnel. By its very nature, there is no room for error.

But what do you do when supervision and case processing can only be carried out by highly educated, highly specialised staff? When adding more full-time equivalents isn’t an option? And how do you reduce risk when parental leave, illness, or staff turnover are realistic everyday scenarios increasing the pressure?

The technology supporting the task

Now imagine that the legacy system is many years old - a common reality for many caseworkers. It’s difficult to navigate. Tiny fields for large amounts of text, frequent copy-pasting into Word documents, and complex merge processes just to get an overview. Many workflows rely on workarounds and deep specialist knowledge, both in terms of occupational safety and IT. A new legacy system is not in scope.
Security - an extra constraint

To add to the complexity: the type and location of radiation sources are details that must never “fly around”. As in not at all. Absolutely not.

It can actually be solved

Until now, this has been something of a headache, but in 2025 new opportunities have emerged: AI-supported case processing.

Luckily, the legacy system can export cases to Excel. And import them back again, which provides an entry point. Also luckily, relevant legislation, guidelines and so forth are already publicly available on the Health Authority’s website. This means we can safely use this information and send it for processing via OpenAI.
And what about radiation source locations? That information remains secure in the browser (on the caseworker’s computer, as always). We reattach the information once we receive the AI-generated response. Then we’re ready to create a supervision report (Word) and sync the case data (Excel).

To ensure clarity about what supervision reports are based on, we’ve built a vector-based database containing only relevant and validated information for case processing. This ensures caseworkers always receive precise and relevant answers. The system is configured to ensure consistency and high quality by drawing exclusively from this curated data pool - never from irrelevant sources.

In addition, we use an agentic workflow that guides caseworkers through process, ensuring that relevant information is structured and handled optimally. When a new guideline appears, it can be added to the data pool with a simple drag and drop.

The user interface is a revolution compared to the existing legacy system’s UI - forms are created on the fly using generative UI, aligned with specialised guidelines so that logos and colours are respected.

It’s still people doing the case processing

A key principle for the solution is that experienced specialists remain in charge. They review - and, if necessary, adjust the conclusions of supervision reports. This isn’t about replacing expertise. It’s about removing repetitive tasks, prepare cases for efficient processing, while at the same time ensuring quality and consistency.
For example, the Health Authority requires that case processing follow precise language guidelines - something far easier with AI than with humans. It also becomes harder to overlook things, reducing the risk of human error significantly.

The solution is about securing quality, timely processing, and not least securing that the workload of scarce resources doesn’t spiral out of control

How do you get started? (estimates)

• Allocate 150 hours to establish a Proof of Concept (POC)
• Allocate 300 hours for a Minimum Viable Product (further development of the POC)
• Allocate 50 hours for GDPR

Do you want to know how we can assist you?

Give us a call or send a message, and we will get in touch with you.