Building Better Math Lessons Together
The central question of this project is clear: how can we use AI to detect errors and misconceptions early on, so students receive targeted feedback and teachers know where the learning process is stagnating? For this project, Lauwers College partners with the Universiteit van Amsterdam and the Eindhoven‑based company Algebrakit, a developer and provider of technology for interactive calculation and mathematics tasks.
Want to learn more about the Smart Math Feedback with AI project?
Teacher in Expert Role
Project leader Thomas Winsemius: “What's great about this project is that the school’s role is fairly large. That’s very positive, because in the end we are building tools that add value for pupils and teachers.”
Co‑creation manager Karmijn Steekelenburg adds: “The school’s needs are central. And the teacher really is in the expert role within the project, and therefore has a major influence on what the prototype will look like. We are very happy with that!”
Step‑by‑Step Learning in MathTabs
The technological basis of the project lies in existing scientific insights, including the so-called SET algorithm (Systematic Error Tracing) from the University of Amsterdam. The prototype is being developed within the web-based application MathTabs, which has been available in beta form since October 2025. In MathTabs students work on mathematics tasks step by step. “The concept is that pupils can work through problems step by step in that digital tool,” says the project leader. This is essential because it is not only the final answer that matters: misconceptions become visible precisely in the intermediate steps. The development goal is a prototype that recognizes systematic errors, provides automatic feedback, guides students according to a consistent strategy, and provides teachers with insight via analyses per student and per class.
From Expert Model to Algorithm
Everything starts in the classroom: in this project, reasoning truly stems from practice. What do you encounter in your school? “You actually start with a kind of expert model of the possible misconceptions that may occur within the topic,” explains co-creation manager Karmijn Steekelenburg. The involved teacher has analyzed a large amount of old exams and student solutions, to feed the system with that data. The project team also gratefully uses the insights of the teacher, who has a very good understanding of common misconceptions due to years of experience. The algorithm then tests which error patterns are structural and which are not. This creates an iterative process in which technology and expertise reinforce each other.
Early Testing in the Classroom
Thomas: “We are working on defining misconceptions in percentage calculations. And we are now at the point where we are moving towards a first version of the tool.” As early as March, a first small-scale pilot will start, probably in two classes. This early testing phase is crucial. “The assumption we are going to test is that the application can help a portion of students along without them needing a teacher or extra external help right away, by signaling common misconceptions early on,” says the project leader. By testing the prototype and collecting data in the classroom early in the project, the team can investigate whether misconceptions are being detected correctly – and whether the model also reveals error patterns that were not yet included in the expert model.
