Lumilo 2 

How might presenting real-time learning analytics to teachers in a mixed-reality device affect teacher helping and studet learning?

Challenge

When a class of students uses AI-based tutoring software, the teacher must monitor the class, make quick judgments as to who needs help beyond what the tutoring software provides, and give one-on-one help efficiently. And then move on to the next student. What makes this task challenging is that students' overt help request often distort the real need for help.

Ongoing Research

In a partnership with Carnegie Learning, Inc., we are scaling Lumilo so teachers can use it for many math topics, without the involvement of researchers or PD staff. We are also creating a recommender system and we are conducting classroom studies to hone the tool and test the effect of key features:  Real time analytics and real-time recommendations.

Project in the news: The aftermath of our 2025 fall pilot 

Highlights

• When teachers use Lumilo, they spend measurably more time with students who have more to learn (as measured by a pre-test) and the class as a whole learns 3x as much, with the students who have more to learn benefiting the most (Holstein et al., 2018).

• Lumilo's teacher-centered design rpoces pioneered the use of "Replay Enactments," a form of advanced prototyping that, by replaying historic student-tutor interaction log data, enables users (teachers in this case) to experience the dynamic behaviors of a new AI-based tool during final iterative refinement cycles (Holstein et al., 2019).

• During classroom pilot studies conducted in the spring and fall fo 2025, in middle-school mathematics classes, Lumilo 2 was well-received by teachers and students, even if they also noted areas for improvement (Jin et al., in press), 

Discoveries

• Preliminary analysis using a reconceptualized version of a prior framework for studying teacher agency indicate that Lumilo sessions trigger teachers' foresightful behavior; which is key for supporting teachers' self-regulatory. A boost in teachers' beliefs on their efficacy to effectively carry out the orchestration appears to boost their sense of productivity. 

• An analysis of teacher interviews and Epistemic Network Analysis (ENA) showed how teachers turn Lumilo indicators into adaptive instructional action. Teachers followed distinct decision-making trajectories across noticing, interpreting, and intervention, enacting quick nudges, sustained instruction, or delayed intervention based on indicators of student idling and struggle.

• An analysis of teachers' self-reported decision-making in classroom situations, supported by real-time Lumilo indicators in scenarios where multiple students require help simultaneously, suggestes that teachers rely on shared decision heuristics. These heuristics reflect trade-offs between providing immediate support, student engagement, and addressing potential learning gaps, highlighting how teachers balance classroom flow while preventing students from falling behind.