FocusGuard
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

FocusGuard is an Android focus coach for students that works from behaviour, not chat. It reads which apps I actually use (Android Usage Access) and knows when I'm inside a study block from my timetable. Every app session is labelled focused, distracted or neutral by a rule engine on the phone that works instantly and offline. Genuinely ambiguous sessions, like YouTube during a DSA block (lecture or rabbit hole?), can go to an optional tier: Meta's open-weight Muse Glimmer running on my own laptop over local Wi-Fi, using the neighbouring apps as context. If it isn't running, the app falls back to rules. When distraction inside a study block crosses my threshold, FocusGuard sends a nudge in the moment, with no internet needed. For tomorrow's plan it builds an anonymized summary (minutes per category, an hourly distraction histogram, timetable and exam countdowns, no app names) and sends only that through a small relay to Muse Spark on the Meta Model API. Muse Spark returns a structured plan: time blocks ordered by exam urgency, scheduled away from my distraction peaks, a focus score and goal, and nudge texts written for my pattern. I can also do a voice check-in ("DBMS for ninety minutes, but I keep checking Instagram"): Muse Voice Transcribe transcribes it and Muse Spark turns it into subject, minutes and blockers. Privacy is enforced in code: raw sessions never leave devices I own, the relay rejects any payload that isn't aggregates with HTTP 422, and a "What leaves my phone?" button shows the exact JSON before it is sent. Built for my world: a timetable, an exam 9 days away, and a phone that turns one YouTube lecture into an hour of Shorts.