NeuraLearn

Project Description

NeuraLearn is a speculative learning system for ADHD college students, built around one idea: protect the productive struggle that real learning requires, and clear only the friction that does not serve the student. Most AI tools do the opposite. They optimize for speed and completion, hand over answers, and quietly remove the cognitive work that learning depends on. For neurodivergent students, who are already failed by a one-size-fits-all education system, that shortcut deepens the gap instead of closing it. The NeuraLearn system centers on an AI agent named Sage. Sage never hands over the answer. It asks, reframes, and re-presents the material, guiding students to their own conclusions while reading their cognitive state to decide when and how to adjust. Built in mixed-reality, this demo simulates cognitive signals rather than live hardware, so people can stand inside the proposal and feel what a system that guides, instead of answering, is actually like. NeuraLearn is research to be continued at Harvard's MDE program, where the work moves toward a closed-loop architecture and a tested system.

Problem

Neurodivergent students are failed by an education system built for an average learner who does not exist. The numbers are stark. Only 58% of neurodiverse students graduate, against 79% of students without disabilities, and just 15% of students with ADHD finish a four-year degree. These students are not failing because they cannot learn. They are failing because the conditions were never built for how they think. Now there is a new pressure. 89% of US college students have used AI to do their schoolwork, and the students the system fails hardest are the ones reaching for the tool that lets them skip the work entirely. Current AI rewards speed, completion, and checking boxes, while learning requires productive struggle, retention, and the ability to apply what you learned. Research from the MIT Media Lab found reduced neural connectivity in people who used AI to write instead of doing it themselves. Offloading physically shows up in the brain, and it damages everybody, especially the students who can least afford it.

Insights

The work moved through several stages: First, secondary research established that the gap was documented rather than personal, drawing on graduation data, AI usage surveys, and neuroscience on cognitive offloading. Second, I defined a design ethos for the agent, Sage, whose toolkit deliberately excludes handing over answers, so it can ask, reframe, and re-present but never solve for the student. Third, I built the spatial prototype in Spline, SwiftUI, and RealityKit, running on Apple Vision Pro with simulated cognitive signals rather than live hardware. Fourth, I engineered a way to capture attention. Apple blocks raw eye gaze on visionOS for privacy, so I reframed the question from where the eyes point to where the head points, polling ARKit head pose at about 60 Hz and processing the logs through a Python pipeline. Fifth, I ran user testing with four participants, two neurotypical and two with ADHD. The strongest insights were that ADHD users wanted agency, not rescue, and that Sage's constant presence could itself pull focus from the task. There was clear need for this kind of system support, but introduced earlier in life to support critical growth stages, milestones, and years of self-blame.

Solutions

To study attention on a platform that blocks eye-tracking, I built a head-pose capture system that produces per-scene heatmaps and trail overlays on the recording. Testing with four participants surfaced clear signals: ADHD users want agency, not rescue, which validated the guide-not-hand-over principle; Sage worked as a body double; and its floating presence and split spatial layout added cognitive load worth redesigning. The honest open question that remains is telling productive struggle from unproductive struggle in real time, which is exactly what the next phase of the research will go after.