Research That Shapes the Product: Netresearch at ICWE 2026 in Lyon

In June, our research team travels to the International Conference on Web Engineering (ICWE 2026) in Lyon. We're bringing a full paper that grew directly out of our work on ScormIQ.

What is ICWE?

ICWE is one of the leading conferences for web architectures and web technologies. This year it takes place June 9 to 12 in Lyon, hosted at INSA Lyon. Proceedings are published by Springer (LNCS), and selected papers get invited for the Journal of Web Engineering.

Out of 39 submissions, the program committee accepted 8 full papers. That's an acceptance rate of 21%. Our paper is one of them.

Our Paper: How Can AI Grade Like a Teacher?

The title: Toward Reliable LLM-Integrated Web Architectures for Teacher-Aligned Automatic Student Grading. The core question: How can a language model grade short answers in a way that matches the expectations of the specific teacher?

The paper describes a modular web architecture with five components. At the center is a feedback loop. Teachers confirm or correct individual grades, and these confirmed cases become training material for all subsequent grading cycles. Cases where the model is uncertain get automatically routed to the teacher. Autonomy grows with evidence.

Results from an evaluation with real course data (Bachelor Computer Science, HTWK Leipzig) show that just 3 confirmed examples are enough to reach an agreement rate above 88% with the tested models. At 10 examples, the best model hits 90%. The key insight isn't which model performs best, but how little feedback is actually needed.

What This Has to Do With ScormIQ

This research isn't an academic exercise. The architecture we describe in the paper forms the foundation for the grading logic in ScormIQ. When a trainer in ScormIQ reviews learner responses, they train the system to match their specific expectations. The paper provides the scientific evidence that this approach works, and describes how the underlying technology needs to be built so that results remain reproducible and verifiable.

For us at Netresearch, that's the difference: we don't just build products with AI. We validate the methodology scientifically. What works in controlled experiments finds its way into our software.

ICWE 2026: June 9 to 12 in Lyon

Our colleague Jonas Gwozdz, who conducts this research as part of his PhD at HTWK Leipzig, will present the paper on site in Lyon. A second paper was also accepted for the PhD Symposium at the same conference.

If you're interested in our work at the intersection of web architectures and AI in education: we'd love to talk. Either at ICWE in Lyon or directly via a ScormIQ demo.