The Atom Learning Model (ALM) tokenises a school curriculum by breaking down math textbooks into individual 'atoms' - specific actions or steps that learners can take. These atoms are linked together with prerequisite relationships, forming a graph structure. The system uses this structure to generate questions and assess children's abilities without human intervention. In a trial deployment in two English secondary schools, the system composed 6,648 questions for 373 children over seven weeks, but its performance was limited by the lack of depth in the question generation. --- Why it matters: This matters because it demonstrates an AI system that can automatically generate educational content and assess student learning without human intervention. This has potential implications for personalized education and large-scale assessment. Source: https://arxiv.org/abs/2608.21106