STEPwise: an LLM generates questions and worksheets to identify where a pupil’s reasoning goes astray in maths

Developed in collaboration with six educational experts, the system breaks down problem-solving into four stages and eight sub-skills, and then generates diagnostic materials for any given problem.

Knowing that a student has made a mistake is easy; understanding when his or her reasoning went wrong requires individual questioning that teachers cannot carry out for an entire class. Researchers and six experts in mathematics didactics have developed the STEP framework, which breaks down problem-solving into four stages and eight diagnostic sub-skills, specifying for each one which clues to collect and evaluate.

This division allows a language model to generate, for any problem written, questions specific to each stage and the corresponding evaluation grids, without the teacher having to design everything by hand. Eight experts judged the pedagogical validity of the material produced, which guided the design of STEPwise, a tool where the teacher retains control over the generation and evaluation.

Tested in class with 5 teachers and 64 students, STEPwise helped each other to locate the breaks in reasoning and encouraged the students to reconsider their approach.