Knowledge-in-a-Box: Offline Course-Grounded AI for Low-Connectivity, Cost-Constrained Classrooms

The crisis

• Billions of people lack reliable connectivity — when the internet goes dark, so does the collective knowledge it carries.
• About 9 in 10 children in sub-Saharan Africa cannot read and understand a simple text by age 10, and an estimated 98 million remain out of school — the only region where that number is still rising.
• Cloud-based AI tutoring assumes steady connectivity and a recurring per-query cost that resource-constrained schools cannot sustain.
• Rural and peri-urban classrooms often run on a shared laptop or a single-board computer, with intermittent power and no dependable internet.

About this research

Most course-grounded AI assumes a reliable network and an ongoing cloud budget — exactly what is missing where help matters most. This thread investigates course-grounded assessment, tutoring, and content generation that run fully offline on low-cost edge hardware, grounded in national curriculum corpora and cited back to the source standards, with the goal of matching the quality of expensive cloud models using small local ones. The organizing constraint is delivering trustworthy, curriculum-aligned support with no runtime cloud dependence in infrastructure- and cost-constrained education settings. It builds directly on the lab's production grading and retrieval work, adapted into an offline, resource-frugal form, and draws on edge LLM inference, structure-aware retrieval, and agentic generation. Faculty-advised.

References

• ITU · Facts and Figures 2024 — global connectivity gap
• World Bank / UNESCO · Learning poverty — children unable to read a simple text by age 10
• UNESCO · Global Education Monitoring Report — Technology in education

Related project