Inspiration Load-shedding kills study sessions across South Africa, and my school network blocks the AI model downloads most study tools depend on. Every AI study tool assumes permanent internet — when the grid fails, studying stops.

What it does GridFail is an offline-first study kit. Snap a matric past-paper photo, get text extracted on-device, tap Explain for a breakdown plus flashcards, then drill yourself with multiple-choice quizzes that surface your weakest topics first. A green OFFLINE pill proves nothing phones home.

How I built it Vite + React + PWA shell, IndexedDB for everything, Tesseract.js for OCR, and a small local model for AI explanations. The key decision: answers appear instantly from an extractive breakdown, then quietly upgrade to AI explanations labeled on-device vs cloud. Unfinished work shows as disabled, never mocked.

Challenges Making on-device AI feel fast on weak hardware and school wifi that blocks model CDNs. Solved with the instant-first tier and honest fallback messaging. Also learned that quiz distractors from other cards' answers test recognition better than random ones.

What's next Full CAPS corpus seed, spaced repetition, and a sub-50kb SMS fallback mode.

Built With

Share this project:

Updates

Submission history