UNIScan — On‑device OCR Scanner

Overview

Uniscan is a privacy-first mobile scanner for Android and iOS (Expo + React Native) that captures documents, cleans them up, recognizes text fully offline, and lets you sign & encrypt files before you share. It ships with two OCR engines: Google ML Kit (unbundled, no Play Services) for Latin/Chinese/Devanagari/Japanese/Korean and Tesseract for Arabic/Cyrillic — both running on the device. A built-in encryption vault (AES-256-GCM) and optional biometric unlock keep clear files short-lived and encrypted by default.

Core flows:

• Scan or import images/PDFs → auto-crop/enhance → OCR on-device → export as text/PDF/DOCX.
• Encrypt before sharing (.uenc envelope) or keep inside the app’s vault.
• Bring files into Uniscan from other apps via Android’s Share sheet (custom native bridge).
• Automatic wipe of decrypted previews, session auto-lock, and screenshot/recording prevention while sensitive screens are open.

Targeted audience

• Field teams & SMBs that need fast capture + OCR with no cloud dependency.
• Legal/finance/healthcare staff who must keep data resident on the device.
• Privacy-conscious individuals who want local-only scanning and encryption.
• Developers/IT needing a secure scanning tool that still plays well with MDM and enterprise sharing.

The challenge

Most mobile scanners either upload to a cloud for OCR or bundle SDKs that phone home. That fails strict compliance and breaks in low-connectivity environments. Teams needed reliable OCR for multiple scripts, strong encryption, and hardened UX (no accidental screenshots or lingering decrypted files) — all without sending data off the device.

Our approach

100% on-device pipeline with native glue where it matters:

• Dual OCR engines, offline:

ML Kit (unbundled) is baked into the app bundle, so recognition runs locally and works without Google Play Services.

Tesseract (Android) via a custom native module (TesseractDirect) for Arabic/Cyrillic and other .traineddata packs. The app includes a tessdata bridge to point Tesseract at bundled language files.

• Defense-in-depth for clear data:

Vault encryption uses AES-256-GCM via react-native-quick-crypto, with keys in SecureStore and optional passphrase + biometrics.

Auto-wipe of temporary decrypted files when the app backgrounds; session auto-lock after inactivity; screen-capture blocking on sensitive screens.

• Share-in / Share-out with control:

Custom ShareBridge (Android) reads incoming share intents so documents can be received privately.

Exports can be encrypted by default (.uenc envelope) before leaving the app, or shared as images/PDF/DOCX when policy allows.

• Predictable file layout (no surprises):

Private app directories such as uni-scans/ (captured images), uni-docs/ (imports/exports), and dedicated vault paths keep clear vs. encrypted data separated and manageable.

The stack

• Framework & runtime: Expo 53, React Native 0.79, React 19, Expo Router.
• Camera & media: expo-camera, expo-image-manipulator, expo-media-library, expo-sharing, expo-document-picker.
• OCR:
• @react-native-ml-kit/text-recognition (unbundled model packs for latin/chinese/devanagari/japanese/korean).
• Tesseract Android native module (TesseractDirect) + tessdata bridge; .traineddata bundled in assets/tessdata/.
• Security & crypto:
• Vault in lib/cryptoVault.ts using AES-256-GCM (react-native-quick-crypto) with PBKDF2 when encrypting via passphrase; keys in Expo SecureStore; optional biometrics via expo-local-authentication.
• Session lock, decrypted file wiper, screen-capture prevention (expo-screen-capture).
• Storage & OS bridges: expo-file-system, AsyncStorage, custom ShareBridge for Android share intents.
• Export & docs: docx, expo-print (PDF), plus image/PDF exporters; optional filename utilities and HTML export helpers.
• UI: Expo vector icons, Safe Area Context, bottom tabs; Skia available for image effects.
• Android config: minSdk 26 / targetSdk 35, ProGuard rules included; ML Kit + Tesseract plugins configure native dependencies at build time.

Results

• Reliable OCR without network for Latin, CJK, Devanagari, Arabic, Cyrillic (and extensible via tessdata).
• Compliance-friendly posture: no analytics SDKs, no background uploads, no third-party cloud OCR.
• Safer sharing with one-tap encryption, short-lived clear previews, and screenshot blocking.
• Smooth UX: tabbed navigation for Scanner, Documents, Editor, Sign & Encrypt, Export, Settings, Vault; device haptics; polished camera flow.