AI Study App — 2026
Flashcard & Quiz Generator
- Role
- Design & Development
- Year
- 2026
- Stack
- Python · Kotlin · LLM APIs · RapidOCR · SM-2
- Code
- View repo ↗
The problem
Students drown in textbooks, and making good flashcards by hand takes hours. Existing generators only skim the opening pages of a long book, so the chapters that actually show up in the exam never become cards — and scanned PDFs defeat most of them entirely.
Key decisions
- 01
Chunked ingestion: sources are split into ~35k-character chunks and mined chunk by chunk, so a 600-page textbook's final chapters get the same coverage as its first pages.
- 02
Local OCR with RapidOCR and PDFium rendering — scanned books and photos are read entirely on-device, with no cloud calls and no page limits.
- 03
MCQs are built with misconception distractors: wrong answers are engineered from common misunderstandings, not random noise, so quizzes diagnose instead of just testing recall.
- 04
Generation runs in a background worker with a refill loop and live progress, so a whole book converts without freezing the UI.
Outcome
Books up to 5,000 pages ingest fully. The app ships three ways from one codebase: a standalone Windows desktop exe, an installable offline-capable PWA, and a native Android build — with Anki .apkg export and SM-2 spaced repetition built in.