The team is creating effective and scalable AI-augmented teachable agents using large language models (LLMs) and state-of-the-art AI technologies. These innovative teachable agents are grounded within a learning-by-teaching (LT) framework, transforming students’ roles from passive learners to proactive teachers. These teachable agents are further orchestrated with multiple learning sciences strategies to engage, motivate, and facilitate students’ math learning.

Project Title

ALTER-Math: AI-augmented Learning by Teaching to Enhance and Renovate Math Learning

  Gainesville, Florida, Nashville, Tennessee
Where it is used:
  • Florida (Orange, Palm Beach, Seminole, Lake, Okaloosa, Nassau, Putnam, Okeechobee, Suwannee, Hardee, Baker, Bradford, Gilchrist, Calhoun, Hamilton, Liberty, and Lafayette Counties)
Project Summary

What’s the problem that the University of Florida is trying to solve?

Doubling the rate of math learning presents significant challenges, such as students’ disengagement and lack of motivation. Many students struggle with grasping mathematical concepts because they can’t relate to the topics and have limited options to tailor their learning experience. These issues disproportionately affect low-income students. Although advancements in AI offer exciting possibilities for the future of education, current applications of AI in education often place students in a passive role, in which they are only receivers of the technology’s decision-making.

What does ALTER-Math do?

ALTER-Math will adopt the learning-by-teaching (LT) framework, which reshapes students’ roles from passive learners into proactive teachers and develops and applies effective and scalable AI-augmented teachable agents using large language models. ALTER-Math will be embedded in Math Nation for algebra learning and influence millions of K-4 students.

What is the wow factor?

While we focus on training LLMs to be better teachers, they are already exceptional students. The learning-by-teaching process perfectly echoes reinforcement learning from human feedback to teach LLMs.

How does it work?

Our solution includes AI scaffolding, such as a teachable agent engineered with personalities to provide a personalized and adaptive learning experience for students working in Math Nation; a mentor agent that empowers students’ teaching with strategic feedback and choices to help them generalize their learning; and an automatic evaluation system to investigate students’ teaching quality and strategies for further reflection and improvement.

Visit or contact Kevin Coulson at

Team Members

Wanli Xing
Associate Professor of Informatics for Education, University of Florida.
Phil Poekert
Director, UF Lastinger Center for Learning, University of Florida.
Zandra de Araujo
Chief Equity Officer and Mathematics Principal, Lastinger Center for Learning, University of Florida.
Gautam Biswas
Cornelius Vanderbilt Professor of Engineering and Professor of Computer Science and Computer Engineering, Vanderbilt University.
Reid Whitaker
Senior Vice President of Research and Measurement, Accelerate Learning.