What separates it from a general-purpose model — building pedagogy into the system architecture itself

1What it takes to call something education AI

A general-purpose conversational model can answer subject questions too. So what distinguishes an AI that calls itself educational? The answer this project offers is that pedagogical knowledge and learning sciences principles are integrated directly into the system architecture. Not handled through clever usage or prompt design, but woven into the design from the start. That difference is what enables context-aware instructional support.

2Four roles for the agents

1. 1Scaffold whole-class discussionPut supports in place so discussion can hold
2. 2Provide individualized tutoringTeach to each student
3. 3Support assessment and feedbackMeasure understanding and return it
4. 4Leverage learning dataPersonalize instruction from what has been collected

Lined up, the four show which part of a teacher work is being taken on: the part that stops scaling as class size grows. The project names three persistent challenges in K-12 education — teacher capacity, uneven access to personalized instruction, and limited opportunity for students to develop AI literacy.

3Who builds it with you

Development proceeds through what is called adaptive co-design, involving not only engineers and researchers but educators, students and school administrators. Educational tools go unused unless they fit classroom practice and deployment constraints. That administrators are included reads as recognition that budget and operational limits decide whether anything gets adopted.

4Measuring in real schools

Parallel efficacy studies evaluate implementation in authentic school settings, examining student engagement in mathematics and science, problem-solving performance, standardized assessment outcomes and cost-effectiveness relative to existing instructional supports. That cost-effectiveness is named explicitly is notable: education technology has to be cheaper than the alternative as well as effective to spread.

Of the NSF awards this site holds as of 2026-09-02, seven are SBIR Phase II, all to companies. Amounts are the obligated amount as of the check date and may change.