Skip to Main Content

SabiqLab

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

SabiqLab — Demo video

I homeschool my daughter while she's still attending external private school. Every week I decide what each child should learn and check their work by hand, and they still feel it as homework. GameSchooling is the tool I built for my own family during this hackathon. A Parent writes one free-text Plan per Child, covering what to learn, in which language, what to avoid, and real-world Rewards with Points prices. Ready-made Starter Plans cover ages 4 to 18. One press of "Make quests" turns the Plan into about five illustrated Quests, with progress streamed live. The Child plays on the Parent's device, with no Child account. They tap a cover, hear the Quest read aloud, and send a Proof. A Proof can be a photo of something in the real world, a voice recording, a typed answer, or a photo of a paper worksheet. Meta models check every Proof. SAM 3.1 finds what the Quest asked for and draws its masks over the photo. Muse Voice Transcribe tells the Parent's prompting apart from the Child's answer. Two independent judges, hosted Muse Spark 1.3 and open-weight Muse Spark Open on Baseten, must both accept with confidence of 0.7 or more before Points are paid. When they are unsure or disagree, the Proof goes to the Parent, who can accept or reject any Verdict. Points reward mastery, not compliance. Stakes are opt-in bets on harder Quests, the Sibling Board is opt-in and age-adjusted under a shared Family Goal, and a weekly Muse Wonder Digest tells the Parent what each Child learned. Every model call is logged with its model, tokens, cost and latency, and shown under each Quest and Verdict. Child names never go to models, audio is deleted right after transcription, and a Guardian check reviews every Quest before a Child sees it. Built solo from an empty repo during the hackathon. Try it at gameschooling.web.id (press Demo or just sign up). Code at https://github.com/ahmadwaliradhi/gameschooling

Team