What goes to the model
Before generating, the app puts together a short brief: which words should appear, what your level is, which topic is chosen and what material is needed. The words come from the ones you are learning right now, or from a specific deck if you point to one. If a grammar topic is set, the material is built around that grammar.
- The list of words: the ones you are learning now, or the deck you picked.
- Your level from settings - A1 to C2.
- The topic: if you set a grammar one, the material is built on it, otherwise a neutral everyday one.
- Your native language, so the translation is written correctly.
What happens next
The model writes the text and the app parses it: it finds your words in it, marks them up for tap-to-translate and cuts the text into phrases so an exercise can be built from them. The finished material is stored on the device.
Why it's different every time
The same set of words yields different texts: another plot, another order, other forms. That isn't a side effect, it's the point - repeating in an identical context memorises the context, not the word.
If something came out wrong
The model does make mistakes: an odd phrase, an unnatural turn, a wrong translation. Write to us from Settings and say what came out wrong - that way we fix the brief for the model, not one story.
