Atlas

Overview

A study site built around one idea: the asset is the interactive scene, not the text. Every post ends in something the reader manipulates — a grid deformed by a matrix, a state machine fed by hand, a diff with the bug inside, a decision with consequences — and that same source file is also the script for the video and the cuts for the shorts. Behind it there is a curriculum of 1598 topics, from first-year university subjects to a senior career path, which doubles as the public roadmap: a topic with a post is linked, one without it shows what is coming.

The Challenge

Most technical explanations online are text with a picture on top: you read them, you nod, and a week later nothing is left. What actually teaches is manipulating the mechanism — moving the matrix and watching the area change sign — and that is expensive to build, so almost nobody does it. The other half of the problem is sustaining it: a site with hundreds of posts written by one person drifts in quality unless the standard is enforced by something that is not memory.

The Approach

The engine came first and knows nothing about the site: scenes are registered by id and loaded on demand, so a post only downloads what it embeds. Anything discrete — arrays, stacks, state machines, code stepping — is DOM and SVG, which stays sharp at any zoom and can be read out loud; anything continuous is canvas. Underneath both, a single model: the algorithm emits a trace of states and the view draws it, which is what makes the teaching testable. The quality standard is a gate: types, linters with no suppressions allowed, tests over the trace of every scene, and content checks that walk all 305 posts on every change.

The Outcome

The site runs in production on my own VPS with its API, its mailer, PostgreSQL and self-hosted analytics, and it holds a full backoffice: curriculum, posts, exercises, students, classes and the audiovisual production board. The private tutoring it supports is the part that pays for the rest.

Key Highlights