A doctor from Nagpur told me something about how she is implementing AI in her clinic that amazed me. She runs a small clinic. 15 years of practice. 40 patients a day. Her biggest problem was never diagnosis. It was follow-up. Patients would come in, get a prescription, and disappear. Half would not complete their medication. Some would not come back for follow-up. A few would show up months later with the same problem, worse. She tried everything. Reminder calls. SMS. A register. Nothing worked. Her assistant forgot. Patients ignored SMS. The register got outdated by Thursday. Then she figured something out. Her patients would not read text messages. But they would listen to a WhatsApp voice note in Marathi. Something about hearing a voice felt personal in a way text did not. She also knew that in many homes the husband controls the wife's phone. So the message had to make sense if someone else opened it. Each patient chose their own follow-up phrase during the first visit. And she knew compliance collapses between day 4 and day 7. Not later. The intervention has to hit on day 4. By day 7 it is too late. She built this using Claude. No code. No developer. WhatsApp voice notes in Marathi on day 4, day 6, and day 10. Personalized. Follow-up compliance went from 40% to over 75%. No engineer could have built this. Not because it is technically hard. Any developer could build it in a weekend. Because no engineer would have known to build it this way. Voice note instead of text. Marathi instead of Hindi. Husband controlling the phone. Custom phrase at first visit. Day 4 intervention window. Every design decision came from 15 years of watching patients in Vidarbha not come back. That is domain knowledge. It does not exist in any dataset or AI course. India produces 1.5 million engineers a year. Huge numbers learning AI. "AI/ML enthusiast." "Prompt engineering certified." A lot of people know how AI works but have no idea how any industry actually works. The engineer builds for the problem they imagine. The domain expert builds for the problem that exists. I have seen this in legal. Engineers built beautiful AI tools for lawyers. Technically impressive. Lawyers do not use them. Because a lawyer's real problem is tracking 150 hearing dates across 4 courts while managing a clerk absent half the time. A CA who filed 500 GST returns knows where errors happen. A teacher who taught 10,000 students knows the problem is not content but attention. No engineer in HSR Layout designs for a student with a shared phone studying between 9 and 11 pm. India has an advantage nobody talks about. Millions of English-speaking domain experts carrying decades of ground-level knowledge. Globally rare. And almost nobody is training them to use AI. India does not have an AI talent shortage. It has domain experts who think they are not "technical enough." They are wrong.