NeuroLoop

Inspiration

Studying alone often means you think you understand something until you try to explain it that's the Feynman Technique in a nutshell. NeuroLoop started as a way to combine that idea with spaced repetition: explain a concept in your own words, get real feedback on where your explanation was weak, and have it resurface right before you'd naturally forget it.

But explaining random topics in isolation only goes so far. Learners usually have a bigger goal in mind "I want to learn React," "I want to understand machine learning" and no sense of what order to tackle it in. We wanted NeuroLoop to help answer that too.

What it does

NeuroLoop now works in three connected steps:

Generate a roadmap : tell it what you want to learn, and it breaks that goal into an ordered sequence of checkpoints, from foundational to advanced. Teach each checkpoint back click into a checkpoint, get a clear explanation of the concept, then write your own explanation of it in your own words (the Feynman Technique). Get real feedback, not just a score: your explanation is evaluated for what you got right, what you missed, and any misconceptions. A "Gap Detector" goes further and pinpoints the exact sentence in your explanation that was vaguest or most likely wrong, so you know precisely what to fix.

Score high enough and the checkpoint is marked mastered, the next one unlocks, and it's scheduled to resurface for review before you're likely to forget it visible in a review queue grouped by "Today," "Tomorrow," and upcoming dates.

**Given the time i had, the demo video shows a strong explanation scoring well and unlocking the next checkpoint.I didn't have time to also stage a deliberately weak explanation to show the lower-score/retry path on camera that path exists and works (a low score keeps the checkpoint in progress so you can retry), it just isn't shown in the recorded demo.

tech stack Frontend/backend: TanStack Start (React 19 + Nitro) AI: Gemini API, called via TanStack server functions Persistence: localStorage — deliberately kept lightweight for a single-user hackathon build rather than standing up a full database

Challenges i ran into Connecting the roadmap to the teach-back flow. Passing the right checkpoint context through navigation, and keeping it in sync even across a page refresh, took more care than expected. Keeping AI output safe to parse. Gemini doesn't always return perfectly formed JSON on the first try; we had to make our parsing forgiving (graceful fallbacks instead of hard failures) so one awkward response doesn't break the whole roadmap. Deciding not to over-build. It was tempting to reach for a full database and calendar UI. We chose localStorage and a grouped list view instead — simpler, faster to ship, and enough for what the app actually needs right now. Accomplishments that we're proud of A genuinely connected loop: generate a roadmap → learn a checkpoint → get specific feedback → unlock the next one → get reminded to review it later — all working end to end, not just as separate demos. The Gap Detector, which goes beyond a generic score to show you the actual sentence that gave away a shaky understanding. Keeping the whole thing lightweight enough to build solo in about a week without sacrificing the core idea. What i learnt Spaced repetition is only motivating if the learner can see what's coming and why a bare "review later" isn't enough; showing "due in 3 days" grouped clearly makes a real difference. Small, forgiving fallbacks in AI-parsing logic matter more than they seem to at first a strict schema quietly turns "AI had an off moment" into "the whole feature is broken." Sequencing matters as much as the feature itself: teaching back a concept in isolation is useful, but a roadmap gives it purpose.

What's next for NeuroLoop Deeper support for procedural/skill-based topics (like coding), where explaining a concept isn't the same as being able to do it likely needs a different evaluation approach than teach-back scoring. Cross-device sync, if NeuroLoop moves beyond a single-device tool. Richer progress visualization a proper calendar view of upcoming reviews, and roadmap-level mastery stats over time.