The first thing you notice about Priya Sharma's classroom in Wardha, Maharashtra, is the noise. Forty-seven students aged 8 to 14, crammed into a room designed for thirty, all talking at once — but not to each other. They're talking to phones.

Each phone runs Priya's creation: a WhatsApp-based AI tutoring system that helps students work through math problems in Marathi, their native language. The AI doesn't give answers. It asks questions. "What do you think happens if you multiply both sides by 3?" "Can you draw this problem?" "What did we learn yesterday that might help here?"

It sounds simple. It wasn't.

The Problem That Started Everything

In 2024, Priya had been teaching for eleven years. She loved it and she was exhausted by it. Rural Indian schools face a well-documented crisis: too many students, too few teachers, vast differences in student ability within a single classroom. Priya's 8-year-olds were learning addition while her 14-year-olds needed algebra. Same room. Same teacher. Same hour.

"I would teach to the middle and lose both ends," she told me over chai in her school's tiny office, a converted storage room with a ceiling fan that works on good days. "The advanced students were bored. The struggling students were lost. I could see it happening and I couldn't fix it."

Priya had heard about ChatGPT on the news. She didn't have a computer. She had a phone — a used Redmi Note, bought for 4,000 rupees (about $50) — and a WhatsApp account.

Building in the Margins

What Priya did next would be unremarkable in a Silicon Valley startup and is extraordinary in context. She taught herself to use the OpenAI API by watching YouTube tutorials on her phone, late at night after her children were asleep. She couldn't read the documentation in English well enough, so she used Google Translate — feeding English docs in, reading Hindi output.

Her first version was crude: a WhatsApp bot that could answer math questions. But she noticed something immediately. When students texted a question to the bot, they engaged differently than when they raised a hand in class. There was no embarrassment. No fear of looking stupid in front of peers. Students who never spoke up in class were sending dozens of messages.

"The phone doesn't judge them," Priya said. "It doesn't get frustrated when they ask the same question five times. I get frustrated. I try not to, but I do. The phone never does."

The Pedagogy Problem

But a bot that answers questions isn't a tutor. Priya's breakthrough was understanding that distinction. She spent three months refining her prompts — she calls them "teaching instructions" — to make the AI behave like a good teacher, not an answer machine.

The AI asks students to explain their reasoning. It offers hints instead of solutions. It connects new problems to ones the student has already solved. It switches between Marathi and Hindi based on the student's responses. When a student is clearly stuck, it simplifies the problem rather than just repeating the explanation.

"I taught the AI how I teach on my best days," Priya said. "The days when I have enough sleep and enough time and only five students to worry about. The AI gives every student my best-day teaching."

Spreading Without a Plan

Priya shared her system with two teacher friends. They shared it with their networks. Within six months, teachers in 40 schools across Maharashtra were using it. Priya hadn't built a company or sought funding. She was just sharing a WhatsApp number.

The system ran on her personal OpenAI account. When the bill hit $200 in one month — a significant sum on a teacher's salary — she posted about it on a local teachers' forum. A tech executive in Mumbai saw the post and offered to cover API costs. Then a nonprofit called the Digital Learning Initiative reached out to formalize the program.

Today, eighteen months after Priya first connected ChatGPT to WhatsApp, her system — now called "Ganit Mitra" (Math Friend) — runs in 200 schools across Maharashtra, Madhya Pradesh, and Karnataka. It handles 15,000 student interactions per day. Preliminary assessments show students using the system improved math scores by 23% compared to control groups.

What Priya Worries About

Success hasn't eliminated Priya's concerns. She worries about dependency: "The AI should make them better at thinking, not replace their thinking." She worries about access: in many villages, not every family has a smartphone, and data costs are real. She worries about quality: "I built this for how I teach. Not every teacher teaches this way. The AI carries my assumptions."

She's also clear-eyed about limits. "AI cannot replace a teacher who sees that a student is hungry, or that something is wrong at home, or that they need a hug more than a math lesson. AI is a tool. It is a very good tool. It is not a teacher."

When I left Wardha, the classroom was still noisy. A 10-year-old girl named Ananya was hunched over a phone, working through a fractions problem. She had sent the AI twelve messages in twenty minutes. On the thirteenth message, she typed: "I understand now!!!!" Three exclamation marks. The AI replied: "Excellent work, Ananya. You figured this out yourself. Can you try a harder one?"

Priya, watching from across the room, smiled.