We delivered an AI Piano Teacher for a Mount Pleasant piano instructor with 38 students and one upright — one that listens to every practice session at home, tells each student exactly which measure went wrong, and writes her a Sunday-night note on who is ready to move on.

She has taught piano out of the front room of her house off Coleman Boulevard since 2009. One Yamaha upright, a bench that has been reupholstered twice, and a waiting list that has not been empty in six years. Thirty-eight students, ages six to seventy-one, each getting thirty minutes a week. The problem was never the thirty minutes.

It was the other 10,050 minutes between lessons. She would send a child home with a Clementi sonatina and a pencil mark on measure 14, and the child would come back the next Tuesday having practiced measure 14 wrong forty times. A wrong fingering, repeated all week, is harder to undo than it was to teach in the first place. She estimated that a third of every lesson was spent unlearning what had been practiced.

She had tried the obvious things. Parents were asked to record practice on their phones and text it over. Some did. She then had 38 students' worth of videos arriving at 9 p.

m. on a school night, and she was watching them on the couch with the sound low while her husband watched the Braves. She stopped after two months. It was not a system.

It was a second job that paid nothing. She came to the Lab on Clements Ferry Road in July with a specific question: could something listen to a student practice at home and hear what she would hear? What was built runs on an ordinary tablet propped on the music stand. The student opens the piece she has assigned, presses record, and plays.

The AI Piano Teacher follows the score in real time. It knows the notes, the rhythm, the fingering she has marked, and the tempo she has set for that student that week. When the student finishes, it tells them, in plain language, what happened. Not a score out of 100.

Something closer to what she would say: you rushed the left hand in measures 9 through 12, the F-sharp in measure 14 was played as F-natural three of the five times, and the ritardando at the end was clean. It does not correct technique it cannot see. It does not pretend to hear wrist tension. It hears pitch, rhythm, dynamics and evenness, which is most of what goes wrong between lessons, and it says so specifically enough that a nine-year-old can go back and fix it.

The part she cares about most arrives Sunday night. A single note, one paragraph per student, covering the week. Who practiced, how many days, what improved, what is still wrong, and a plain recommendation: ready to move on, needs one more week, or bring this to Tuesday's lesson because it is a hands-together problem. She reads it in twenty minutes and walks into Monday knowing what every lesson is for.

The numbers moved faster than she expected. In the first six weeks, the average number of practice days per student went from 2. 8 to 4. 6 a week, because a child who gets told immediately that measure 14 is fixed will play it again.

The portion of lesson time spent on unlearning dropped to what she calls "almost nothing," and she has started assigning repertoire a full grade level sooner for eleven of her students. Two adult beginners who had been stalled for a year on the same Bach minuet finished it in a month. She also stopped losing students. Her attrition had run at about six a year, almost all of them children who quietly stopped practicing and then quietly stopped coming.

Since June she has lost none, and she believes it is because the practice itself became less lonely. Someone is listening. It tells them when they get it right. Her recital in May will be at a church on Pitt Street, as it has been for fifteen years.

She expects it to be the first one where she is not nervous for anybody. "I used to teach thirty minutes a week and hope. Now I teach thirty minutes a week and know."