Artificial intelligence technology such as generative models and agents are key to finally automating away some of financial services’ biggest data challenges. Developing these solutions requires expertise, vision and a commitment to quality.
So, what are we at Duco doing in this space? We caught up with Kaja Zupanc to find out. Kaja is our Head of Machine learning, leading our team of experts in Antwerp, Belgium. She was also recently named Waters Technology’s Data Science Professional of the Year (Vendor) in the Women in Technology and Data Awards 2025.
She told us about the latest research with Flanders Innovation & Entrepreneurship (VLAIO) she and her team are doing to keep Duco’s customers at the forefront of AI technology.
Q: Why are you working with VLAIO?
Kaja: VLAIO is a Flemish government organisation which funds research with the aim of stimulating growth and innovation in the region. Not all the companies they work with are Belgian, but their primary focus is on keeping innovation and knowledge within the country. They help innovative companies to fund research and development that pushes the boundaries of their chosen field.
Duco’s machine learning team first started working with VLAIO as part of Metamaze, which Duco acquired in 2024. We’ve just completed our third project with them, having expanded its scope to leverage the broader capabilities of the Duco platform. At this point, we've built up substantial experience working together with VLAIO.
Q: What projects have we worked on with VLAIO support? What’s our latest project?
Kaja: We always focus on developing something new: we start with a baseline of existing technology and we develop a new approach that should significantly outperform the baseline. Our previous project focused on reducing human effort while increasing the speed of machine learning models for Intelligent Document Processing (IDP). Our latest project is on integrating generative AI into our adaptive Intelligent Document Processing (AIDP) solution.
This was before genAI became such a hot topic for the industry, but we recognised the potential of the technology a while back and wanted to explore how it could work in practice.
Our end goal was to determine where this technology would be most effective. For example, as part of the project we also proved that accuracy for the main tasks of document processing was lower with zero-shot and few-shot genAI compared to our in-house supervised models.
(Few-shot learning is where the model is only trained on a small number of annotated documents. Zero-shot learning is where the model hasn’t been specifically trained on any examples at all.)
However, when it came to the onboarding process of our customers specifically, we were able to significantly speed this up by enabling zero-shot and few-shot genAI models. So it’s always about adding the right kind of AI for the right task to ensure you’re generating value. A lot of people are very keen to get the latest AI trend, but that’s not necessarily going to help them.
As part of the project we’ve also developed a multi-agent solution to integrate into the Duco platform. An agent is an autonomous system that can make decisions and take actions independently to achieve goals.
Q: That’s a hot topic right now, but when did this project start?
Kaja: This was two years ago, long before the wider industry buzz around it. The reason why it got accepted so fast for VLAIO support is because at that moment no one else was doing it. At the time we decided to do it, it was a really futuristic concept.
It's so crazy that that is what we are working on at the moment! It was a big decision – as I said, no one else was working on it back then. If we were going down the wrong route we were about to lose a whole year of research and development. But taking risks is part of being an innovative company.
And now, where other companies are starting to look into genAI and agents because they’ve become buzzwords, we already have years of research and proven results under our belts.
Q: What’s the response to the project been like?
Kaja: We always get really good feedback from VLAIO. One of the reasons that they fund our projects is that, even though we focus on development, we document everything really well. We are one of the few that can show things in production, such as performing a live demo.
We also contacted our VLAIO project advisor, Veerle Desmet, for feedback on our project:
“This project was definitely one of the first that I’ve seen bringing GenAI into functional business applications as part of the VLAIO-funded project.”
Thanks Kaja! We also discussed some very exciting things coming up in the near future. We can’t talk about them just yet, but trust us – you’ll want to keep an eye out for some announcements in the coming months.
In the meantime, why not discover how we leverage generative AI in our adaptive Intelligent Document Processing solution?
