Project Info
This project did not submit a demo video on Devpost.
Inspiration
In Afghanistan, girls above age 11 are completely banned from school. This isn’t just a case of poor education access—it’s the total removal of it. It is the only country in the world with such a restriction. As of now, 1.4 million girls have been affected. In response, many kids rely on underground homeschool hubs, where they might share a single textbook across dozens and are taught by underqualified, impromptu teachers doing their best in impossible conditions. While AI has reshaped education globally, its benefits haven’t reached communities like these. Most EdTech tools rely on constant internet, use only English, and are too expensive to scale — leaving behind the very students who need them most.
What it does
So we built Sage — a local AI network where students connect to a single offline PC that serves as a powerful AI tutor for an entire class of 30–50. Despite no cloud, no internet, and limited hardware, our AI tutor delivers high-functioning features like (1) support for underrepresented languages, (2) step-by-step questions and hints, and (3) a teacher dashboard that tracks which concepts students struggle the most with.
How we built it
One small PC emits a completely offline Wi-Fi local server. Students connect to it through their phones or tablets. They open a browser and gain access to a full AI-powered tutoring platform.
Challenges we ran into
When integrating Simon's Jupyter notebook LLM model into our codebase to connect the backend and frontend, there were a lot of compatability issues. The LLM was running on the GPU, but the full-stack application ran on the CPU. Fine-tuning the model was a challenge in order to improve the accuracy of English to any foreign language translation. That, and the Mistral model's code was not fully compatible with the Flask back-end. Another challenge were merge conflicts with the User Interface completed by David and Abdur.
Accomplishments we're proud of
There were 2 beginner teammates, Ren and Simon, participating in the hackathon for their first time. Ren did a great job with pitching and coming up with the slide deck for judge presentations. Simon did a great job with training LLM models on the GPU to get the theoretical AI translation for different language. The other two experienced teammates, Abdur and David, worked hard to ensure the UI was consistent with the hackathon idea. We are proud of developing a innovative idea that has great business value for future work.
What we learned
We learned that for integrating LLM models, it's important to check the CPU/GPU compatibility. Doing so would speed up the integration process. Our developers got better at: -Figuring out how to model our data -Coming up with sleek UI Designs -Training enough data for accurate offline LLM functionalities Troubleshooting Intel NUC hardware
What's next
We plan to add more foreign language translations. Additionally, we plan to train more data for additional educational curriculum topics that would support Science, English, and Programming.
This project did not link a GitHub repository.
Analysis
No indexed repository for this project, so there are no commit stats to show.
Technology
- FlaskUnchecked
- Mistral AIUnchecked
- PythonUnchecked
- PyTorchUnchecked
- ReactUnchecked
- TypeScriptUnchecked
No repository was indexed for this project, so these Devpost claims have not been checked against code.
AI coding agents
No repository was indexed, so agent usage could not be checked.
Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.
Codebase size
No repository was indexed, so there is no codebase to measure.
This project did not link a GitHub repository, so there is nothing to diagram.
This project did not link a GitHub repository, so its feature claims have not been checked against code.
Export this project's context (description, README, evidence, key source files) to chat with an AI agent elsewhere.