Data science consultancy in Arnhem

It builds models that predict outcomes, such as the load on a section of the grid or the number of hospital admissions, and keeps them working in day-to-day operations. These models are part of the AI applications we build.

If you’re looking for advice first on where AI can add value in your organisation, see AI consultant Arnhem. That covers the choices you need to make; this page focuses on the model itself and on data science consultancy in Gelderland.

No. LUMC is based in Leiden. We mention this case study as an example of our work with models in production in a clinical setting. We haven’t published a case study involving an organisation in Gelderland.

Our office is on Westerdoksdijk in Amsterdam. In Arnhem, Nijmegen and the rest of Gelderland, we work on site or online.

cairelab is LUMC’s centre of expertise for AI in healthcare. LUMC wanted not only to develop and implement AI, but also to manage it sustainably in day-to-day care, using on-premises, sovereign infrastructure.

We worked on model development, experiment tracking, CI/CD, model serving, monitoring and maintenance, as well as integration with clinical systems. The case study also mentions MLflow, Docker, FHIR/HL7 and DICOM/PACS. Read the case study.

Arnhem is home to TenneT’s Dutch headquarters and Alliander’s registered office. Liander, part of Alliander, manages regional gas and electricity networks in Gelderland and elsewhere.

A model could predict the load on a section of the grid, help plan maintenance or detect anomalies in meter readings early. Such models use time-series data that reflects seasonal and weather patterns. Read more about our work in the sector at data and AI for the energy sector.

By monitoring its inputs and predictions. With data drift, the model’s inputs change, for example because of new meters or a different way of recording data. With concept drift, the underlying relationship changes, such as when a neighbourhood adopts solar panels on a large scale.

The actual outcome often only becomes available later. So you first measure shifts in the inputs, then compare predictions with outcomes. Read how we set this up under MLOps.

Often through FHIR, an HL7 standard for exchanging healthcare information electronically. The model reads data from the electronic health record (EHR) and writes its output back to a system the healthcare professional uses. Alongside FHIR/HL7, the LUMC case study mentions DICOM/PACS for medical images.

The region is home to Radboudumc in Nijmegen and Rijnstate in Arnhem, among others. Software that provides information for diagnostic or therapeutic decisions may fall under the MDR; your regulatory team assesses this.

Not always, but it can. Patient data and data about critical infrastructure are subject to strict requirements on where they can be stored. In that case, the model runs in your own environment, with the same components as in the cloud: version control, a model registry, automated deployment and monitoring.

At LUMC, the emphasis was on on-premises, sovereign infrastructure to safeguard security, privacy and control over data and models. You make that choice in advance because it determines which tools you can use.

If a fixed rule, an established score or a good report already answers the question, or if the outcome isn’t recorded anywhere. Start with the data or a dashboard instead. Read about how we build models for images, text and predictions under machine learning.

For predictions about individual patients or customers, with an emphasis on validation, read data science consultancy in Utrecht.