AI Learning Box

The Project

AI Learning Box turns one small, relatively low-cost Raspberry Pi 5 into an offline learning server. Students may connect to it from a tablet, phone, or laptop on a local network and get two things out of it: An AI tutor that they can chat with to ask questions and get explanations A library of structured lessons and storybooks, including storybooks in Kinyarwanda, the students’ own language. No internet is needed, and no student data ever leaves the device. I built it with three open-source tools on the Raspberry Pi: Ollama: Allows me to run an open-source AI model (gemma2:2b) directly on the device. Open-WebUI: Gives the AI a simple chat interface in the web browser. Kolibri: Helps provide offline lessons and storybooks, such as Khan Academy and African Storybook (which has Kinyarwanda storybooks) through the Raspberry Pi server. Hardware: Raspberry Pi 5, 8GB RAM, 256GB microSD card, in a case with a cooling fan. It runs headless (no screen) as a local server. AI model: Gemma2:2b, run with Ollama. I chose it because it is small enough to answer quickly on a Raspberry Pi without any serious issues. For better Kinyarwanda support I am also testing aya-expanse:8b, a multilingual model. What was hard: The Raspberry Pi has limited memory, so I had to pick a small, fast model without compromising too much on functionality. I also had to make Open-WebUI (in a Docker container) connect to Ollama, fix the Kolibri data folder setup, and find an AI model that can handle Kinyarwanda. Since most small models do this poorly, this was probably the hardest step throughout the entire process. What I learned: I learned more about the Linux command line, Docker, and how large language models run locally. I also learned about "low-resource languages" like Kinyarwanda that most AI models are not good at. Additionally, I got to find more about the limits of running AI on limited hardware. What's next: I am adding a Wi-Fi access point mode so the Raspberry Pi makes its own Wi-Fi network so that it will work in a school with no internet at all. Why it matters: In the age where AI is being implemented everywhere, to not have access to such technology will most likely lead to lagging behind those that integrate AI into various things. With this, schools without internet connections can still give its students an AI learning experience and a digital library, not just in English but also in their own language, for the price of one small computer.

Ai

About the team

  • South Korea

Team members

  • Hyunjun