MATHS 130 is a large first year mathematics course, and generative AI was making take home assignments less useful as evidence of learning. At the same time, students could often produce a correct answer without really understanding the reasoning behind it, so a written assignment showed very little about how they were thinking.

Instead of trying to make the assignment AI proof, Tanya Evans changed what is assessed. Students record a video of two to five minutes in which they explain their solution, why each step is valid and how the steps connect, often simply by pointing a phone at their handwritten work.

FeedbackFruits then shares each video with three or four classmates, who assess the correctness and clarity of the reasoning with a structured rubric. Students record through Panopto inside Canvas, subtitles are added automatically and peer marks go straight back into Canvas, so the design works at the scale of a large course.

The peer assessed part is deliberately low stakes: 40% for submitting the video, 30% for completing the peer reviews and 30% for the average mark from peers. Students may use AI to explore the mathematics, but they have to explain the reasoning themselves, which makes their thinking visible and assessable.

Compared with a semester before ChatGPT with the same teaching team, the pass rate rose by 16%, exam absenteeism dropped from more than 15% to 9% and more students earned A and B grades. The team sees this as an encouraging result that further semesters will need to confirm.