Teaching a Billion People to Code: How JupyterLite Is Scaling the Impossible

Picture half a million French high school students learning math and programming in their individual browsers, all running simultaneously from a single server. For Sylvain Corlay, founder and CEO of QuantStack and member of the Jupyter Steering Committee, this is not a pie-in-the-sky stretch goal. Corlay and his team helped make it happen by building JupyterLite, a serverless distribution of JupyterLab running entirely in the browser — kernel included, no Docker necessary. In this On the Road episode of The New Stack Makers, Corlay sat down with TNS Editor in Chief Heather Joslyn at JupyterCon in San Diego to discuss how an “accidental startup” evolved into a 30-person team pushing the boundaries of what’s possible with browser-based computation and why that matters for the future of worldwide technical accessibility. From Quant to Quantum Leap QuantStack began in 2016 as what Corlay calls “almost an accidental startup” —  a one-person company meant to support his work as a Jupyter core developer after relocating from his job as a quantitative analyst at Bloomberg, in New York, to France. The timing was perfect: Jupyter had exploded in adoption, but most maintainers were in academia or large corporations, creating huge demand for consulting services. Today, QuantStack is embedded in multiple kernel ecosystems beyond Jupyter, including Conda-Forge and Apache Arrow. But it’s JupyterLite (started by QuantStack team member Jeremy Tuloup in 2021) that represents the organization’s most wide-ranging mission. “JupyterLite brings a level of scalability that Jupyter had never had before,” Corlay said, pointing to the French school system deployment, called Capytale, as proof of concept: half a million high school students registered, over 200,000 user sessions per week, all essentially served from one server. “All there is on this server is a content management system for the teaching material, the notebooks; everything else is served on CDNs, and all of the code is executed in the end user’s browser,” he said. “You can do simple math and Python programming in the browser very effectively — and you don’t need to spawn a Docker image in the cloud at scale for millions of users and require an elastic deployment of JupyterHub.” The Billion-User Challenge But half a million users is just the beginning. Corlay’s vision extends to countries with younger, larger populations. “If you look at Nigeria, with a population of 210 million forecast to grow significantly, all of these kids are going to go to school and, because it’s the 21st century, presumably learn how to code,” he said. “It’s not reasonable to expect each user session to run in a Docker image somewhere. Nigeria probably doesn’t have infrastructure for that scale.” JupyterLite’s low-tech approach, however — essentially a static page served from a CDN —  makes it feasible, and on a global scale. “I really think that we are on to something that could be used to teach programming to a billion people,” he said. Making that vision into reality, though, requires addressing challenges beyond pure compute: meeting full accessibility standards for users with disabilities, real-time collaboration and expanding QuantStack’s open source Emscripten-Forge software distribution to provide more packages for browser-based execution. It’s work that corporate clients don’t typically fund, requiring grants and foundation support. Check out the full episode to hear more about QuantStack’s origins at Bloomberg, the company’s sustainable growth philosophy, and hints about an upcoming announcement about the convergence of “all of the things that people from various parts of the stack have worked on,” into something Corlay describes as “a large crossover thing” that’s worth staying tuned for. The post Teaching a Billion People to Code: How JupyterLite Is Scaling the Impossible appeared first on The New Stack.