Skip to content
Book Open access

ManipulativeAgent: A Multi-Agent System for Generating Interactive Mathematical Manipulatives

Nov 2026 · Proceedings of the 35th ACM International Conference on Information and Knowledge Management · 0 citations · 15 references

Abstract

Virtual interactive manipulatives help elementary students transition from concrete to abstract mathematical thinking through hands-on digital activities. However, creating them requires both pedagogical and programming expertise, while existing platforms only offer fixed templates that cannot adapt to individual teaching needs. We present ManipulativeAgent, a multi-agent system that allows teachers to create customized manipulatives from natural language descriptions of their instructional goals, with human-in-the-loop checkpoints for review and revision at each stage. Rather than relying on a single large language model (LLM) for end-to-end generation, ManipulativeAgent distributes the task across specialized agents that collaborate through a shared component library of pre-tested virtual tools. This design separates pedagogical reasoning from code implementation, enabling each agent to focus on its strength while the component library ensures interaction reliability. Evaluation on 40 real teaching scenarios shows improvements over commercial platforms and general-purpose LLMs, with the largest gain in interaction richness. A user study with 89 frontline teachers from 15 provinces shows that 92.1% would continue using the system for lesson preparation. A demonstration video is available at https://youtu.be/7eclhXhRy0A.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.