AFRICA DEEP TECH CHALLENGE 2025 — TECHNICAL PROJECT REPORT
Offline-First AI-Powered Learning System for Resource-Constrained Environments
Across Sub-Saharan Africa more than 240 million learners live in bandwidth-poor regions where the cost of one video lecture can exceed a family’s weekly income. Schools lack up-to-date laboratories, and lesson delivery is mostly one-to-many, leaving little room for personalized feedback. The result is low STEM attainment and an ever-widening skills gap.
The challenge is clear:
| Constraint | Real-World Impact |
|---|---|
| Intermittent / no connectivity | 58 % of Nigerian students learn entirely offline outside school hours |
| High data costs | Average 1 GB mobile data = 27 % of daily minimum wage |
| Low-end devices | 60 % of student devices == 2 GB RAM |
| Power outages | 70 % schools experience < 6 h electricity / day |
To break this cycle we must push intelligence to the edge—making advanced, interactive learning possible without constant internet, cloud GPUs or large batteries.
We built Skoolio: an offline-first learning ecosystem that fits inside any Android or low-spec desktop device while delivering:
Unlike streaming platforms, Skoolio requires 0 MB of data after an initial 1.5 GB bundle download and runs for 12 h on a 3000 mAh battery.
| Category | Constraint | Design Response | |
|---|---|---|---|
| Power | ≤ 5 W typical Android phone | All inference under 900 mW; heavy ops use WebGPU burst then sleep | |
| Memory | 1–2 GB RAM | 4-bit quantization, streaming weights; upper bound 1.5GB | |
| Compute | Cortex-A53 @1.2 GHz single core common | Models distilled to 1 B params, ~3.2 token/s offline | |
| Connectivity | Offline > 90 % of session | CDN-less — content sideload via SD card, Bluetooth mesh, or 1-time hotspot | |
| Data | Sensitive student scores | All analytics synced only on user opt-in |