DIREC supports VELUX and Grundfos in the race for smarter AI

Large data centers and massive energy consumption. These are currently necessities for today’s mainstream AI tools. But it doesn’t have to be that way. Researchers and companies are now working to develop advanced, embedded algorithms – known as embedded AI – that allows devices to make autonomous decisions without being connected to the internet.

Pumps and skylights are obvious candidates for this new development. That’s why VELUX and Grundfos share a common ambition: to harness embedded AI to optimize energy efficiency, enable predictive maintenance, and improve indoor climate.

“We often work with extremely resource-constrained systems that lack file systems or internet access. That’s why we are focused on identifying the most efficient algorithms to improve our products. It essential that we evaluate different solutions in a structured way,” says Thorkild Kvisgaard, Head of Electronics at Grundfos.

Grundfos and VELUX are part of the DIREC project “Embedded AI” together with leading researchers at DTU. Here, Amin Hasanpour, a PhD student at DTU Compute, has developed EdgeMark – a benchmarking tool designed to compare and evaluate the various tools in the field.

“Many companies struggle to choose the right tools for their products due to the sheer number of options and the complexity of factors to consider. I’m happy that my research can help companies make informed decisions,” says Amin Hasanpour, who spent a year developing EdgeMark.

EdgeMark impresses with just one click


Globally, embedded AI research is attracting increasing attention, and the number of researchers in the field is growing rapidly. EdgeMark represents a major step forward in making this technology accessible to the business sector, and the tool has already impressed VELUX:

“EdgeMark can significantly shorten our time-to-market. It automates critical insights about algorithm performance, meaning we don’t need a full team of developers to achieve the same outcomes,” says Glenn Berg Andreasen, Control Specialist at VELUX.

According to Xenofon Fafoutis, project manager and associate professor at DTU Compute, embedded AI is a rapidly evolving field because it challenges the prevailing dependence on energy-hungry, cloud-based systems that often compromise data security.

“Embedded AI forces us to think more efficiently because the algorithms must operate in environments with limited computing power. As we improve our ability to design lightweight efficient algorithms, we can help companies create better products while reducing CO₂ emissions, improving security, and protecting privacy,” he explains.

From solo pioneers to collaboration and synergy


The collaboration between DTU, Grundfos, and VELUX has not only provided the companies with a valuable tool for future product development. It has also laid the groundwork for broader commercial cooperation. Grundfos and VELUX are currently testing a joint use case involving embedded AI.

“Investing in cutting-edge AI is a significant and challenging task. Collaborating with Grundfos has been especially valuable, as they face many of the same challenges as VELUX,” says Glenn Berg Andreasen.

Thorkild Kvisgaard from Grundfos agrees:

“In large, complex projects, input from other companies and researchers is crucial. It increases the likelihood that we make the right decisions when selecting technologies and suppliers for the future.”

DIREC brings together experts and strengthens industry

Danish researchers and companies are playing a major role in the global development of embedded AI. Through initiatives like EdgeMark, DIREC is building bridges between academia and industry – creating a collaborative ecosystem that strengthens innovation, efficiency, and Denmark’s international influence.

If you want to learn more about research in this area, you can read our feature about Emil Njor’s passion for embedded AI. You can also visit our project page to learn more about embedded AI.