ProductiveMath: AI-Enabled Small-Group Math Learning
This project advances ProductiveMath, an AI-powered platform designed to help teachers create, evaluate, and facilitate Productive Failure mathematics experiences in small groups.
A central goal of the project is to improve the quality of AI-generated instructional content. Using Evidence-Centered Design (ECD), the project will develop AI evaluators that can help teachers determine whether mathematics problems are accurate, instructionally appropriate, aligned with standards and classroom resources, and well designed for Productive Failure.
The project will also enhance context-aware problem generation and develop teacher-in-the-loop approaches to dynamic student grouping. Together, these tools are intended to support teachers in creating challenging but accessible learning experiences in which students can explore multiple strategies, learn from initial failure, collaborate with peers, and develop deeper mathematical understanding.
Project Description
Project Plan
The project will proceed through several interconnected phases:
Conduct participatory design, usability studies, teacher pilots, and classroom-based studies to iteratively improve the tools.
Develop an ECD-informed framework and rubric for evaluating Productive Failure mathematics problems.
Examine and compare existing AI evaluation approaches with a ProductiveMath-specific evaluator.
Develop and test an AI learning-linter that provides formative and summative feedback to teachers as they create and revise problems.
Design a chat-based workflow that allows teachers to interpret evaluator feedback and iteratively improve their problems.
Enhance ProductiveMath to generate problems based on classroom context, available resources, student readiness, and teacher goals.
Develop a teacher-in-the-loop dynamic grouping system that recommends student groups and explains its recommendations.
Project Details
- Principal Investigator: Seyedahmad Rahimi, Ph.D.
- Institution: Georgia Institute of Technology
- Project Duration: October 2026 – October 2027
- Funder: Gates Foundation
- Project Amount: $499,498
Broader Goal
The project sits at the intersection of AI infrastructure, learning sciences, and product research and development. In addition to advancing ProductiveMath, the project will produce reusable evaluator frameworks, design principles, research findings, and technical resources that can inform the development of other AI-enabled educational tools.
Ultimately, the project seeks to explore how AI can support—not replace—teacher expertise while creating learning environments where students become more adaptive, collaborative, and creative problem solvers who can learn productively from failure.
Research Questions
1. AI Evaluation of Productive Failure Problems
How can AI evaluators be designed and validated to provide useful and actionable feedback on the quality of mathematics problems designed for Productive Failure?
2. Context-Aware Problem Generation and Dynamic Grouping
How can information about students, classroom resources, instructional goals, and learning context be used to generate more appropriate Productive Failure problems and support effective dynamic student grouping?