Using multi-agent system and evidence-centered design to integrate educator expertise within generated feedback
Generative AI (GenAI) can support mathematics teacher learning by automating parts of teacher rehearsals, or opportunities for teachers to practice specific mathematics teaching skills, receive feedback, and refine their approaches. However, little is known about how to design GenAI solutions that incorporate the expertise of mathematics teacher educators into the design of GenAI solutions. This article proposes a three-part design framework: first, by building on the modeling logic from evidence-centered design (ECD) to develop design templates that capture the complexity of mathematics teaching; second, by translating the expertise from these templates into prompts for GenAI; and third, by sharing how a multi-agent system that uses GenAI to simulate teacher rehearsals and feedback can keep the complexity of that expertise by splitting it across agents. We explore how this framework could serve as a general approach for designing simulations—both in its theoretical foundation and technical implementation—that support complex, interactive rehearsals intended to enhance teacher learning.