Yes — this is core to how Leo gets better. As users correct transcriptions in the app, those corrections feed back into training, so the model's accuracy compounds across successive releases. (There's no per-user custom model: improvements benefit everyone.)
