Atlas.
A learning platform that will not pass you until you have earned it. Strict grading, mastery gates, and a tutor that only teaches what it knows.
IN PLAIN ENGLISH
What this is.
atlas is a learning platform that will not let a student move on until they have earned it. Lessons, quizzes, mastery gates, and a retrieval-only tutor make the learning loop explicit.
01 / THE PROBLEM
Progress bars can hide that nothing stuck.
Most self-teaching tools reward motion. atlas uses a ten-topic course about how language models work, from tokens and embeddings through retrieval, and makes each topic prove itself before the next opens.
02 / HOW IT WORKS
The model grades; code decides.
Multiple-choice answers are deterministic. Written answers go to a strict grader with the question, rubric, and answer only. A score of 80 percent or better opens the next topic and schedules review. The model never decides access, and its score is clamped in code.
03 / ENGINEERING DECISIONS
Keep judgment narrow and mechanics pure.
Mastery math, placement, scheduling, and progression are pure functions with no model or database dependency. The tutor retrieves only chunks from atlas lessons and cites them. If nothing relevant is found, it says the course does not cover the question instead of guessing.
04 / WHERE IT STANDS
Build-ready with a deliberate single-user scope.
atlas is wired for Next.js 16 and Neon Postgres with 79 tests across the learning loop, grader, and retrieval. A golden set of twelve labeled answers guards the cheap grading model against over-crediting vague answers. A stronger grader can be swapped in later; the seam exists, the provider is not wired yet.
THE HONEST VERSION
What I am actually counting.
The live demo is the clearest way to feel the mastery loop. The test suite is the evidence for the harder claim: mechanical progression stays mechanical.
TRY IT
Try the interaction.
Interactive demo loading. The case-study explanation remains available without JavaScript.
TOOLS AND CONSTRAINTS
- Next.js 16
- Neon Postgres
- Vector retrieval
- Deterministic grading rules
- Golden-set evals