Shipping to 200K students

My second internship started with a deadline that didn't really make sense. Estudo Play, a Paraíba-based EdTech, had a contract on the table with the Secretaria de Educação de Minas Gerais — a statewide rollout to public high schools preparing for the Enem. The web platform was ready. The mobile app, which the contract hinged on, didn't exist yet. Two weeks to ship.

This was early 2024. ChatGPT had been out a year. Nobody was correcting handwritten essays at state-network scale with AI yet — not in Portuguese, not from cell-phone photos, not aligned to the Enem's five-competency rubric. The pitch was that students would keep writing essays the way they always had — by hand, in class — photograph them with the app, and get a graded report with feedback by competence within minutes. The web team and the AI team had built that pipeline before us. Now we needed the front door.

Four of us were put on the app. None of us had shipped real mobile software before. We picked React Native — at least the language was familiar — and started building, learning the platform as we went, taking reviews from senior engineers between sprints, killing scope ruthlessly. I led the frontend and app development: design system, screen flows, the integration layer that talked to the essay-correction and personalized study-path APIs the AI team was building in parallel.

I also designed the app in parallel — most of the screens went through Figma before they ever became code. The Dribbble shot from that work is here: dribbble.com/shots/24375026-Mobile-App-Design.

Two weeks later the app was in stores. The contract closed for over R$37 million. Minas Gerais rolled it out across the state network starting April 2024, targeting 3rd-year high schoolers and EJA students preparing for the Enem, and the state launched a companion event called Esquenta Enem MG with live classes for the first simulado. Within months, Rio de Janeiro and Bahia followed.

The numbers from the first stretch: 1M+ accesses, 600K+ essays corrected by AI, 750K printed test notebooks, 85% adherence among enrolled students. By the end of that run, the app reached 2,000+ public schools and 200K+ students across the three states.

Then in September–October 2025, the platform did something nobody else had done: 461,100 handwritten essays corrected by AI in a single month — a Guinness World Record for the largest volume of AI-graded handwritten essays ever recorded. It made national press (Época Negócios) and local coverage in Paraíba (Portal Correio).

The thing I keep coming back to: the Enem is the gateway to public university in Brazil, and public schools have historically not been able to afford essay correction at the scale you'd need to actually prepare students for it. Bringing that cost to near-zero, while keeping the rubric aligned to the real exam, is the kind of leverage that's hard to design around. None of the pieces were exotic on their own — OCR, an LLM pipeline aligned to a public rubric, a React Native shell. The trick was shipping all of it together, fast enough to win the contract that put it in front of 200K students.