Client Overview

A research group at Georgia Tech studies how students learn math and physics. Its work depends on two pieces of software: a web application that helps teachers create math problems with AI, and a physics learning game that students play in the browser. Both are used in studies with real teachers and students.

The Challenge

Research software is usually built by students and rotating contributors, one study at a time. By the time a platform has to run in classrooms, it needs things a prototype does not have: one maintained codebase, reliable sign-in and data storage, a record of what each student did, and a team that can keep shipping between studies.

Our Solution

Smplfy took on both platforms as the group’s engineering team, working from the researchers’ priorities in regular review meetings.

An AI problem builder for teachers

Teachers step through a guided form: grade level, difficulty, length, the standard they are teaching, what students already know, and a theme their class cares about. The application generates a story problem to match, with images and a teacher support package, and saves it to a shared library where it can be edited, regenerated with an instruction, or exported as a PDF. Teachers used it to create problems in a classroom study.

In progress: saved classrooms with their available materials, so a generated problem can include a hands-on activity with what is actually in the room, and a chat assistant that fills in the form from a plain-language description. Student names and notes are never sent to the AI model.

A physics learning game with study-grade logging

The game and its level editor had grown as two separate codebases. We merged them into one, unified the physics behaviour, and added a hub where a user’s role decides whether they play levels or build them. A cloud backend handles accounts and saved work.

For the research itself, we replaced the previous learning-record service with an open-source record store and dashboard that captures what students do in the game. An in-game AI assistant that students can ask for help is in development.

Engineering practice that survives the grant

Both platforms are getting the groundwork research code often lacks: automated tests for the main teacher workflows, written documentation, and a ticketed backlog that the researchers can read.

Results So Far

  • Teachers created problems with the problem builder during a classroom study.
  • The physics game and level editor now ship from one codebase.
  • Student activity in the game is captured in a record store the group controls.
  • Both engagements are ongoing.

Conclusion

A research group’s time should go to research. With an engineering team that carries both platforms from one study to the next, the researchers can plan studies around what they want to learn, not around what the software can survive.