GenAI in Education
Generative AI can support education when it is designed around pedagogy, learner agency, teacher workflows, and meaningful evaluation. My work focuses on how GenAI systems can provide useful feedback, support practice, and adapt to learners without replacing human judgment or reducing learning to automated scoring.
Feedback on open-ended answers
This project explored how AI can support the analysis of students’ written answers by processing open-ended responses and helping identify patterns in understanding, misconceptions, and answer quality.
Problem
This project aims to develop an AI service that provides personalized feedback and formative assessment on students' answers to open-ended questions in the context of high school or university level education. Although modern AI can be prompted to provide such feedback or assessment using effective example-based strategies and context engineering, the results are neither consistent not always appropriate for the context.
HIGHLIGHT
Gap
Traditional automated grading usually focuses on correctness, while manual qualitative analysis is slow and difficult to scale. Many systems also fail to show why an answer is weak, what concept may be missing, or how answer patterns differ across students. This limits their usefulness for formative feedback and classroom decision-making.
Approach
Nehir, a Masters student developed a system for processing students’ written answers and extracting meaningful information from them. The system was designed to support analysis of student responses by organizing answers, identifying relevant patterns, and making open-ended response data easier to interpret for research or teaching purposes.
Results
At this point, a comprehensive grading module is complete and has been evaluated against four datasets. We found that the our approach matched human grading more in the datasets that required detailed reasoning or step by step explanation rather than the ones that require a more objective answer. The module on feedback generation is under development, and is planned to be integrated with an advanced course on Evolutionary Biology in February 2027.
The project demonstrated how AI-supported answer processing can help move beyond simple right/wrong scoring. It created a foundation for analyzing student understanding more systematically and for supporting teachers or researchers in identifying patterns that would otherwise remain hidden in large sets of written responses.
If you are a Masters student at Utrecht University, and are interested in taking this project to the next stage, you can connect with me and share a brief statement indicating your motivation to work on this project.
Generating content is so 2025'
The value of GenAI depends less on whether a model can produce explanations, summaries, or feedback, and more on whether students can understand, trust, and act on that output. A useful GenAI system should support reflection, practice, and teacher guidance while making its limitations visible.