THE SOUND IS OLDER THAN THE MEANING. BEFORE YOU can begin to understand a verse of the Bhagavatam, it is a shape in the air—long syllable, short syllable, an ancient metre. It has passed from mouth to mouth for as long as anyone has recited it. Prathosh AP, a 37-year-old assistant professor of machine learning at the Indian Institute of Science (IISc) in Bengaluru, decided to pass it through a machine, some 18,000 verses of it, chanted as a pandit would, in a voice that would never tire. During his ongoing semester break, Prathosh created and published a tool he called Vagdhenu, which can render any Sanskrit verse as parayana, the recitation that a pandit would give. “The response has been pleasant. I’ve got two million page hits worldwide. I didn’t expect this.” The model has been downloaded 1,500 times from the repository where he posted it. “People are not just using it for chanting,” he says, “but also for brushing up on conversational Sanskrit”. He is now trying to build a feedback system. “Imagine if you could chant a shloka to the tool and it could help you correct your mistakes.” Prathosh set out to preserve traditional chanting and has ended up, more or less by accident, building a tutor.