Two-Minute Speaking is built on one idea: speaking improves when you produce sentences and get a precise correction, repeatedly, on the things you personally get wrong. Here is exactly what happens inside the app and why.
The first session sets your level
You answer a few short prompts by voice or text. From that, the app estimates your CEFR level and builds a first picture of how you speak: sentence length, the structures you avoid, the sounds that need work. This takes a couple of minutes and is never a test you can fail.
Every session is a scenario from your life
The app keeps a life profile: your job, the people you talk to, your routines, the places you go, and what you want English for. Sessions are generated from that profile, so instead of “order a coffee in London” you get “explain to your manager why the report is late” or “tell your neighbour about the leak”, whatever is actually yours. The prompt arrives in your native language so you know precisely what to say; the answer has to be in English, out loud.
One correction, one lesson
After you answer, the coach shows what you said, the native-like version, and teaches exactly one point, explained in your language. One, not five. A learner who is told five things remembers none of them; a learner who is told one thing, and then meets it again two days later, keeps it.
The error ledger
Every mistake is recorded as a named, specific item in one of six categories: grammar, vocabulary, nativeness, pronunciation, writing, and discourse. Not “you make grammar mistakes” but “you drop the article before singular countable nouns”. You can open the ledger and read it. So can the coach, which is how the next session knows what to bring back.
The mastery rule
An item closes only after you use it correctly in N separate sessions with spaced intervals, scheduled by FSRS, a modern spaced-repetition algorithm. N is 3 by default and you can set it anywhere from 1 to 10. Getting something right once, in the session where you were just told it, proves nothing; getting it right on Tuesday, Friday and the next Wednesday, without being reminded, does.
Never more than three open
At most three items are open at any time. While three problems are open, the app does not introduce new territory. This is the rule people notice first and like least, and then like most: the first week feels slow, and after that your speech stops repeating the same errors.
A coverage map, not a syllabus
Behind the ledger is a map of English concepts, levelled by CEFR, each marked not started, introduced, practising or mastered. It is not a course you march through. It is a map of what you have already shown you can do, and what has not come up yet.
Your schedule, your variant
You choose how many sessions a day and at what times; reminders follow that. You choose American or British English as the target. Coaching is in Persian or English today, with other languages generated on demand.
What it runs on
The app is a progressive web app at app.aienglishspeaking.app: open it in your phone’s browser and add it to the home screen. Voice goes through streaming speech recognition and text-to-speech; the coaching itself is generated by Claude models with structured outputs, so corrections are consistent, not improvised. Native iOS and Android apps are planned.
What it is not
It is not a chatbot to talk to for an hour. It is not a grammar course. It does not promise a band score. It is a daily two-minute correction loop that knows you, and it is honest about the fact that speaking gets better slowly and then all at once.