what is llm-academy fr?

LLM Academy is a plugin for Claude Code, the command line coding assistant, that teaches people how large language models actually work. Rather than following a single fixed course, it starts with a short assessment that figures out what a learner already knows and builds a path around the gaps, skipping material already mastered. The academy teaches a widely recommended set of resources for learning about AI: 3Blue1Brown's Neural Networks video series for visual intuition, Andrej Karpathy's Neural Networks: Zero to Hero for building things from scratch in code, Dwarkesh Patel's chalkboard lectures on how frontier models are trained and served, and Raghav Dixit's short piece on vectors. Instead of simply pointing to these sources, the academy explains each concept in its own words, gates progress behind requiring the learner to explain ideas back, and links out to the original material for full credit. There are two learning tracks. The explainer track covers understanding without writing any code, aimed at people who have never programmed. The builder track adds all eight of Karpathy's code along lectures, where the learner's own code is actually run and graded. Across both tracks there are 22 units organized into 6 tiers, moving from the basic idea that everything becomes numbers up through attention, transformers, and how a model like GPT stores facts, ending in a final exam that asks the learner to explain how ChatGPT works end to end in their own words. To keep knowledge from fading, the system includes mastery checks that require an 85 percent score to pass, spaced review sessions timed to when a learner would normally start forgetting, plus experience points, ranks, and streaks to track progress. The academy is also designed to grow: when new source material is added, a command slots it into the learner's existing path. Installation happens through Claude Code's plugin system, adding the marketplace and installing the plugin, then restarting and running a start command. The project is licensed under MIT, though the videos and articles it teaches from remain the property of their original authors.