Our first voice build was text-to-speech over the existing lesson. Students hated it, politely, by never using it twice.
Interruption is the feature
In a real session the student cuts in at the exact moment they lose the thread. If the tutor finishes its paragraph first, that moment is gone. So barge-in had to be instant. The tutor stops mid-word and the board holds at the current stroke.
The board can't pause
Silence with a frozen board feels broken. While the tutor listens, the board keeps its last step highlighted and the cursor breathing, so the session still feels alive.
Latency budget
- Under 300ms to acknowledge that you've started speaking
- Under 900ms to the first word of the reply
- Under 3s to the first new stroke on the board
Want to see this in practice?
The board is the whole argument. Watch a lesson, or join the waitlist and bring the topic you're stuck on.