Shadhra

Readers of printed books lose the lines that moved them: in phone photos, scattered notes, and margins they never reopen. Shadhra keeps them in one place.

Live on the web at shadhra.app and on Google Play.

What it does

• Capture: photograph a book page, select the passage, and the text is extracted automatically (Arabic or English), then reviewed and saved with its book, author, and page.
• Reflect and organize: notes on every quote, books, authors, tags, collections, and links between quotes.
• Find: full-text and meaning-based search, plus a chat that answers from your own library.
• Remember: a daily quote from your archive and "on this day".
• Share: quote cards and short essays built from your quotes.

How it is built

• Backend: FastAPI (Python 3.12), PostgreSQL with Alembic migrations, Redis and a background worker for OCR and AI jobs.
• Clients: React + Vite + Tailwind web app, and an Expo / React Native Android app on one shared API.
• AI layer: a multi-provider router with fallbacks, per-feature cost tracking, and spending limits.
• Quality gates: CI runs linting, type checks, about 300 backend and web test files, and an OCR accuracy gate that fails the build if the character error rate on a fixed Arabic test set exceeds 5%.
• Operations: an admin console with feature flags, kill switches, and usage dashboards.

What I learned

Building the product was the easy half. Early user research showed most new users never saved a first quote, so the latest releases focus on one thing: getting a reader from install to first saved quote in under a minute.