dhipeople
Expert Data & Modelling and Gen AI Lead, Denmark

Driving innovation with machine learning and artificial intelligence

Clemens Cremer began his journey with our German (Hamburg) office, where he collaborated across multiple teams and supported projects in various areas including water resources, ports, marine and offshore wind. Three years ago, he moved to HQ in Denmark, where he actively bridges the gap between research, innovation and commercial applications. Drawing from his project experience, Clemens has been at the forefront of our Machine Learning and AI initiatives during this time, focusing on engineering applications and AI Assistants (LLMs). Get to know Clemens and learn more about the innovative work he’s driving.


What was your dream job growing up?

 

I am not sure I ever had a dream job – sure, the five-year-old me might have said firefighter or something alike – but overall, I think I’ve always defined myself more by interests and curiosity rather than by a single job. I think this mindset might originate from me growing up consciously in the 1990s and 2000s, a time when the rise of affordable personal computers and the internet made it particularly evident how fast job requirements, demands and job titles can change. 


In terms of interest, these times might have sparked my curiosity in all things new and digital. Mixing in some interest in environmental challenges and physics, extending to their linkage to societal, political and economic questions. This combination of interests led me to study computational engineering (a.k.a. putting physics into computers) and eventually become an advisor in the physics, water and digital sectors – a good fit overall that allows me to engage with complex challenges while sharing knowledge with others.

 

Please explain artificial intelligence (AI) and machine learning (ML) to a 5-year-old! 

 

Hmm, this is different from what I typically get asked by our clients. Let’s try with a somewhat Christmas-time relevant example of candles:


Imagine your parents telling you: ‘Don’t touch the flame of a lit candle; it’s hot.’ If you listen, then you have learned from your parents’ advice. This is also one way AI can learn – from examples where someone has already linked things together, like 'candle = hot'. We call this supervised learning, because like a parent supervising a child, someone helps the AI learn by providing labelled examples.


If, on the other hand, nobody has told you, and you accidentally touch the flame, you feel that it is hot and hurts. The next time you see a flame, you remember this. This is another way how AI learns – from experience, or in computer terms: data! This we call unsupervised learning.


So, you can see – AI can learn similar to us. One cool thing is that AI can learn from thousands of experiences super quickly and without getting hurt.

 

Could you give us examples of how DHI is using AI/ML technologies to address environmental challenges? 

 

Alright, let’s just think along a classical engineering workflow and thinking along the fast-good-cheap dimensions of consulting. 

 

The typical workflow might start with data gathering and validation. Here, for instance the WANDA (Water ANomaly Detection Application) project demonstrates how AI can make all three dimensions possible. It automatically validates complex environmental sensor data, making the process faster and more reliable than manual checking, while being cost-effective for water utilities and environmental agencies.

 

For data processing and classification, the technology deployed e.g. in projects like GreenUp uses computer vision for urban land cover classification, while MUSE AI detects and classifies birds near wind farms. These applications make previously time-consuming manual tasks both faster – and thereby quick decisions feasible – and more accurate, especially when processing large amounts of data.

 

In operational forecasting, our Venice storm surge prediction system shows how AI can enhance traditional simulation approaches. By combining physics-based understanding with machine learning, we achieve both speed and accuracy in critical decision-making scenarios. 

 

Other approaches we deploy might also combine physics based models and machine learning as – during inference – fast surrogates, or for increased accuracy in hybrid models.

 

Finally, we're using Large Language Models to democratise data analysis and database interaction. This makes complex environmental data accessible to a broader range of users, not just data scientists, effectively balancing all three dimensions: fast access, good quality insights and cost-effective analysis.

 

You’ve also been on DHI’s Global Mobility Scholarship and spent two months in Australia in 2024. Tell us about your experience!


Travelling is always enriching, bringing new perspectives, and in this instance, having enough time to truly ‘arrive’ and get immersed over some months really added to the experience. Of course, it also feels particularly meaningful to return to in-person connections and travelling after the COVID pandemic.


The extended stay provided me with a deeper understanding of our Australian business and clients across very different sectors. For instance, being immersed in the Australian maritime sector was particularly insightful, as it provided another practical perspective on how ports outside of Europe use our digital services in their daily operations, e.g., to optimise their capacity and reduce delays.


A challenge was posed for me to manage the time difference between Australia and Europe – where I still had colleagues and clients to communicate with. I think this again sharpened my view on the intricacies of global collaboration. 

And lastly, if you could have a superpower, what would it be? 

 

Just from a learning and curiosity perspective I would say an impeccable memory and perfect recall would come in handy to absorb and retain more. I would also finally be able to remember names at networking events.

'AI can learn similar to us - both from being taught and from direct experience, but with one cool advantage: it can learn from thousands of experiences super quickly.'

Clemens Cremer

Expert Data & Modelling and GenAI Lead, Denmark

Join our world of water