When LessonLoop entered the Tools Competition, the team had a clear mission: help teachers understand and improve student engagement in real time. What they didn’t expect was that the same work would push them to build an AI-powered system to detect cultural, socioeconomic, and accessibility bias in classroom lessons.
Today, LessonLoop is not only a leader in real-time engagement assessment, but also one of the first ed tech platforms to operationalize bias mitigation in a way teachers can use immediately. As founder and CEO Nona Ullman put it, “Until you start, you don’t know what’s possible until you build something. And then you see, you learn.”
Bias in instruction isn’t always intentional. It often comes from the assumptions teachers unknowingly make when constructing activities, examples, or language. Ullman describes the issue simply: students want to “be seen and heard for who they are.” Yet most lesson templates and materials reflect a narrow cultural perspective.
Before winning a Tools prize in 2024, LessonLoop’s focus was clear: create a way for teachers to measure student engagement at the end of each lesson and then generate evidence-based strategies to improve it.
“What gets measured gets taught,” Ullman said. “We believe that in order to have more engagement in the classroom, you need a way to measure it in real time, and then you need to know how to improve it.”
But the team quickly discovered that generating an activity and measuring engagement was only part of the challenge. The platform’s output needed to reflect the diversity, lived experiences, and needs of actual classrooms.
“We didn’t realize that we had to counter bias,” Ullman said. “We just built it, and then realized it was not bringing in… our values. So then you have to go back and ask, how do we make sure that it reflects our values?”
That realization sparked the creation of LessonLoop’s “Blind Spot Checker,” an AI-powered bias detection system embedded directly into the lesson-planning workflow.
Beyond cultural bias, LessonLoop identified two additional blind spots:
- Language bias that subtly excludes students based on gender, ability, or socioeconomic background
- Accessibility bias that overlooks learning exceptionalities or assumes all students can access content the same way
In short: without deliberate design, AI-generated lessons risk replicating the same inequities teachers have long struggled with.
LessonLoop’s approach was intentionally human-first. The company assembled a diverse team, including a reviewer with lived experience across multiple identities, to examine every output and identify gaps. They spent months defining what “bias” actually means in the context of a classroom lesson.
“The first step was identifying how we define bias,” Ullman said. “We spent a lot of time defining that.”
From there, they built a high-level prompt that screens each lesson for cultural references, tone, and accessibility, followed by a rubric to test whether the revised lesson aligns with defined values.
LessonLoop tested 50 different lessons, comparing AI outputs with and without the Blind Spot Checker. The patterns were clear.
“The biggest gap was mentioning different cultures,” Ullman noted. “Language is generally pretty inclusive… but cultural inclusiveness, that’s just not generally built into lessons.”
The Blind Spot Checker is simple by design. After a teacher generates a lesson activity, they can click a button to run it through the bias review engine. The team discovered that combining generation and bias detection in a single step degraded the quality of both outputs. The two-step design ensures clarity, speed, and accuracy.
LessonLoop’s experience holds several lessons for the broader edtech community:
- Bias mitigation is not a plug-in; it’s a design philosophy.
- Testing is as important as building.
- AI amplifies whatever foundation you build, including bias.
- Tools Competition funding enables risk-taking.
This innovation has already earned recognition: LessonLoop is one of only two companies nationally certified by Digital Promise for prioritizing racial equity in AI design.
LessonLoop continues to refine its approach, exploring learner profiles, accessibility scaffolds, and broader applications of their bias-mitigation framework. As Ullman puts it, “We’re five years in… and yet we’re just at the very beginning of the journey.”
