As the head of Microsoft’s AI for Good Lab, Juan Lavista Ferres and his researchers can spend months building real-world AI solutions. Fifth graders that he teaches built an accessibility tool in their class.
The students in a computer science class at the Global Idea School, an independent, non-profit elementary school in Redmond, Wash., learned vibe coding through GitHub Spark and built a Braille 3D Generator, a tool that turns text into printable, tactile 3D Braille models in seconds.
“We live in an amazing time,” said Lavista Ferres, a 17-year Microsoft vet who has taught at the school, co-founded by his wife, for seven years. “The fact that a 10-year-old can do it in a class without any training? That thing is an actual working solution.”
Six students worked on the Braille 3D Generator. They were inspired by the idea of creating signage to help blind or low-vision people navigate in their school to find classrooms.
The group is the youngest to enter the “AI for a Better World” competition, a national initiative in collaboration with MIT, that invites students in grades 6-12 to explore how artificial intelligence can improve their communities and the broader world.
The students interviewed Anne Taylor, principal program manager for Microsoft Accessibility and an expert in Braille embossers that convert digital text into raised Braille text on paper. Taylor was able to provide feedback and help the students fine-tune their solution so that it was useful for someone who is blind.
The students also visited Microsoft’s Inclusive Tech Lab where they saw how people interact with specialized computer keyboards, game controllers and more.
“I think it would be very good to help people with disabilities,” said Grayson, 10, one of the students in the class. “We’re trying to help the people who can’t see with this Braille project to make it more affordable, so they can tell areas easier — because it would be cheaper for areas to have Braille instead of having to go through a really expensive process.”
For the students, the process felt unlike anything they had done before in class, where they had previously used block-based coding tools like Code.org.
“Instead of having to type the code, we could just say English to the AI and it would make this whole app,” Grayson said.
Vibe coding is a style of software development in which the programmer describes what they want in plain English and lets AI generate the underlying code. GitHub Spark, a tool from Microsoft-owned GitHub, takes that approach and lets users build and deploy web applications through natural language prompts alone — no coding experience required.
What surprised even Lavista Ferres was the leap from browser-based app to physical object. GitHub Spark typically generates React code for web applications, and he didn’t realize it could produce 3D models until the students tried it. Some of their early attempts didn’t work, but they kept experimenting.
“When I saw the output, I was like, ‘wow,’” he said. “I’ve been vibe coding for some time now. I wasn’t aware that we could do this.”
Lavista Ferres started at Microsoft as a data scientist in 2009 and became a lab director 10 years later. The AI for Good Lab operates as part of Microsoft Philanthropies, separate from the company’s product groups. Last year the lab launched an AI for Good Open Call to support projects in public health, education, sustainability, and humanitarian action.
Lavista Ferres said the kids in his class could be future Microsoft colleagues, because they’re creating real-world applied solutions that work.
“This is a new world,” he said. “I show it to all my team and say, ‘Guys, if these kids can do this, you guys can be much more productive. We need to start using this technology more and more.’”
