Interview a wetland
There is no shipped curriculum. You drop in your own material—reading transcripts, photos of book pages, exam papers—and the system extracts what you don't know yet, writes the lessons, and grades what you produce.
Raw context in, curriculum out
An LLM-in-the-loop cycle reads your material against your current level per skill, then writes lessons, drills, and cards where the gaps actually are. A graded lexical corpus double-checks the classifier.
Speaking and listening are first-class
ElevenLabs voices generate dictation and shadowing audio; your recorded speech is transcribed and scored word-by-word against the target. A separate judge model reviews free production—and you can override it.
No streak, no flame, no nag
Spaced repetition with a retention dial, a capped new-card budget, and an honest "you're caught up" ending. Progress is quiet data—a due forecast and mature-word counts, not gamification.
Répétez :« Les mécanismes de financement doivent être vérifiables. »
Language study is the simplest thing Anima does, simulating live conversations at a user's chosen level.