DUET
Diagrammatic UML & ER Tutor: upload a diagram, hand-drawn or digital, and get reflective feedback while there is still time to revise it.
UML class diagrams and entity-relationship (ER) diagrams are foundational in computing, but they ask learners to think abstractly and to master both syntax and semantics. Assessing them means interpreting each student’s individual design, so in large courses feedback often arrives only after grading, reduced to brief, rubric-oriented comments.
DUET, the Diagrammatic UML & ER Tutor, gives formative feedback while students are still working. Students upload a diagram as an image, hand-drawn or digital. A multimodal language model pipeline converts it to PlantUML, compares its structure with an instructor’s reference solution, and phrases the differences as neutral, reflective prompts that encourage revision without revealing the solution. Students can rate each message and resubmit, and educators manage assignments and editable reference solutions.
Interviews with educators shaped the design. In an introductory object-oriented programming course, 135 students then voluntarily uploaded 530 UML class diagrams while preparing for a graded modelling assignment. Half revised their diagram at least once, typically within minutes of receiving feedback, and 74% of submissions received feedback based on a faithful conversion of the drawing. Students found the feedback supportive and manageable but often doubted its correctness, while their confidence in class diagram modelling tended to increase. The team has released the tool, its prompts, and the study data.
People
Papers
- Sebastian Gürtl, Naaz Sibia, David Kerschbaumer, Christian Guetl, Michael Liut, Alexander Steinmaurer."DUET: Rapid Automated Feedback for UML Class Diagrams and ER Diagrams" Proceedings of the 26th Koli Calling International Conference on Computing Education Research. (2026).[doi][]
- Sebastian Gürtl, Gloria Schimetta, David Kerschbaumer, Michael Liut, Alexander Steinmaurer."Automated Feedback on Student-Generated UML and ER Diagrams Using Large Language Models" ArXiv. (2025).[doi][]



