Every child is a sky filling with stars.
Atlas remembers what a child has truly mastered across 1,590 micro-topics, watches each memory fade, and tells you, or your AI tutor, exactly what to light up next.
Chart the sky
A gentle placement, a few questions per five-minute session, finds where the child really is. A right answer vouches for everything underneath it; a miss sends the next question one step down.
Teach at the frontier
Each question is picked live: refresh what is about to fade, then open the next topic within reach. A miss makes that star weaker and sends Atlas to check the building block underneath.
Keep stars bright
Memory fades, so every star has a half-life. Stars go sleepy when recall is at risk, and harder work quietly re-brightens everything it stands on.
The whole tutoring loop is three calls. Your LLM stays thin; the graph carries the pedagogy.
# what to ask right now: placement probe, refresh, or new topic
GET /api/v1/learners/:id/session/next
# record what happened; credit flows down the graph,
# a miss weakens the node and flags the shaky prerequisite
POST /api/v1/learners/:id/session/answer { "topicId": "…", "grade": "correct" }
# the full memory state: strength, recall, due dates
GET /api/v1/learners/:id/memory