Danish startup Teton has used Gefion, one of Europe’s most powerful supercomputers, to turbocharge the development of its AI “care companion” for hospital staff. 

Teton installs cameras and sensors in hospital rooms to gather real-time data. This gets fed to an AI algorithm, creating a virtual “digital twin” of the room. The model monitors patient and staff behaviour such as movement, breathing, or posture changes. If it sees a problem, the system alerts nurses via an app. 

To protect privacy, all processing takes place on-device and none is sent to the cloud. No personal data or raw video footage is stored. Instead, the video is immediately converted into 3D spatial representations and the original footage is deleted. 

Teton’s goal is to help hospital staff spot early signs of health issues and respond faster to potential risks. The company also aims to ease the workload of nurses. By continuously monitoring patient rooms, the system could reduce the number of manual checks. It can also automatically generate care notes. 

The AI care companion has already been trialled at two Danish hospitals with promising results. The tool cut nightshift workloads by up to 25%, according to the company’s own research. 

Teton’s founders, Mikkel Wad Thorsen and Esben Klint Thorius, have spent years training their AI model on real patient scenarios. They even hired actors and nurses to repeatedly act out scenes to improve their training data. However, these manual processes take a lot of time. That’s where Gefion, Denmark’s AI supercomputer, came in.

Using the supercomputer, Teton was able to create over 1 million 3D scenes of hospital rooms. In each virtual room, a different scenario would play out. For example, a patient turning over in bed, a nurse checking vitals, or someone experiencing distress. 

These scenes aren’t just static images, they’re dynamic environments that mimic how people and objects move and behave in real life. By generating so many of them, Teton can expose its AI to a huge variety of scenarios — far more than would be feasible manually. This helps the system learn how to recognise subtle patterns and respond appropriately in real clinical settings.

“Gefion has made a huge difference to the rate at which we have been able to develop our AI model,” Thorsen told TNW.

“Being able to unlock the geometric knowledge helps us to build in much better predictions and estimations of how a patient is doing,” he added. 

Thorsen hopes the improvements will ultimately help hospitals deliver better patient care — and take the weight off care staff.

Giving nurses a helping hand

The EU currently has a shortage of 1.2 million doctors, nurses, and midwives. There is also a declining interest in nursing careers across more than half of the bloc’s 27 countries, according to OECD research.

The nursing shortage is a worldwide issue, which has given rise to several startups trying to fill the gap with tech solutions. One of them is Stockholm-based Bemlo, which has created a platform for short-term hiring of nurses and doctors, allowing hospitals to quickly onboard new staff. Other efforts look to assist nurses more directly. US-based Diligent Robotics has built Moxi, a metre-tall mobile robot designed to help alleviate boring and repetitive tasks in healthcare facilities.

Armed with supercomputing power and digital twins, Teton is betting AI can help ease the strain on stretched hospital employees. And now the startup is looking to fast-track the development of its algorithm. “With all of these new data points, we could significantly accelerate the iteration time and unlock a new scale for the next generation of our model,” said Thorsen.