AI plays a carefully bounded role in our educational application, “Mojo”. Mojo is an AI-enabled assistant teacher that supports differentiated instruction for students through rigorous learning activities aligned to their district’s chosen curriculum. Unlike systems where AI is tasked with generating content directly to students, our approach uses AI primarily behind the scenes to categorize student responses to Mojo’s text-dependent questions, enabling more personalized and effective learning experiences in a two-part process.

Every student interaction with Mojo follows a consistent two-part process:

Part 1: Behind-the-Scenes Classification When a student responds to an open-ended question, the OpenAI GPT-4o model analyzes their answer against specific, human authored criteria developed for that question. The AI categorizes the response based on whether it meets all, some, or none of the question criteria (i.e. learning objectives). This classification happens entirely “under the hood” and is not student-facing.

Part 2: Structured Feedback Delivery Once categorized, the student receives feedback consisting of:

• A brief acknowledgment that mirrors back their thinking (AI-assisted active listening)
• A pre-authored “back pocket” question specifically designed for their response category (human-authored)

Think of it like a sophisticated sorting system that helps route students to the most helpful next question based on their current understanding, rather than an AI generating educational content on its own.

With our small team of less than 20 staff members, we also leverage AI to support our content authoring process. AI assists our curriculum experts in:

• Drafting initial activity frameworks that align with specific curriculum standards
• Generating potential criteria for evaluating student responses
• Suggesting possible back-pocket questions for human review and refinement
• Streamlining the review process to ensure content quality and alignment

This application of AI in our workflow allows our small team to develop truly curriculum-aligned content for Mojo at a scale that would otherwise be impossible, ensuring that participating teachers have Mojo activities that support their curriculum implementations by providing critical instructional differentiation and support to their classroom of diverse learners.

Our expert educators:

• Oversee the design and selection of all instructional questions;
• Develop specific criteria for evaluating responses; and
• Develop “back pocket” questions for each possible response category to ensure Mojo activities are curriculum aligned and pedagogically sound. 
• They also set up and carefully test the guardrails for how the AI categorizes responses, ensuring that each student interaction with Mojo takes place within a “walled garden” guaranteeing that each interaction is narrowly focused on the district’s curriculum.

These carefully designed, tightly bounded applications of AI provide several important educational benefits:

For Students:

• More immediate feedback than traditionally possible
• Content-specific support tailored to their demonstrated understanding
• Increased engagement through responsive questioning

For Teachers:

• Unprecedented visibility into student thinking patterns
• Data to inform individual, small group, and whole-class instruction
• Alerts when students provide inappropriate responses

By combining thoughtful human expertise with carefully bounded AI capabilities, we’re creating a learning environment in Mojo where technology enhances rather than replaces the critical human relationships at the heart of education. Our unwavering commitment is to ensure all students have equitable access to high-quality instructional materials and the personalized support they need to engage deeply with challenging academic content. We believe this balanced approach represents the best of what educational technology can offer—amplifying great teaching rather than attempting to substitute for it.