react-native-loom-example

This repo contains:

• loom/: the Loom engine source tree (incl. the Welvet C-ABI bridge)
• 0_75/: prebuilt Welvet (Loom C-ABI) mobile binaries for v0.75.x
  • 0_75/android_arm64/welvet.so (+ welvet.h)
  • 0_75/ios_arm64/welvet.a (+ welvet.h)
• loom-expo/: an Expo + React Native demo app that runs a small Loom model on-device and stores events in SQLite

The goal is a realistic demo for a diabetes scenario:

“I took meds late, went out, and after walking around (e.g. Costco) my blood sugar can drop. It’d be great to have something that flags risk locally.”

---

What the app does

The app in loom-expo/ implements a simple “risk score” flow:

• You enter:
  • Blood glucose (mg/dL)
  • Minutes since meds
  • Walking intensity (steps/min)
• The app computes a risk score (0..1):
  • If Loom is available (dev client / built APK): it uses Loom/Welvet on-device
  • If Loom is not available (e.g. Expo Go): it uses a deterministic fallback heuristic
• Each “check” is written to SQLite on-device via expo-sqlite
• The Insights screen shows a simple trend chart using react-native-gifted-charts
• A demo notification reminder can be scheduled via expo-notifications
• A demo PIN is stored via expo-secure-store
• A WebView screen demonstrates in-app web rendering

This is an example project and not medical advice. It’s structured as a foundation you can extend with real signals (CGM ingestion, step sensors, insulin-on-board estimation, personal baselines, etc.).

Demo screenshot

---

Tech stack (as implemented)

• Expo (dev client) + React Native
• TypeScript
• expo-router (routes live under loom-expo/app)
• expo-sqlite (local SQLite storage)
• nativewind + tailwindcss (Tailwind-style className utilities)
• react-native-gifted-charts + react-native-svg (charts)
• expo-notifications (reminders)
• expo-secure-store (secure key/value)
• @react-native-community/datetimepicker (time picker in Settings)
• react-native-webview (web screen)

---

Repo layout

App

• loom-expo/app/
  • _layout.tsx: router layout + NativeWind CSS import
  • index.tsx: Dashboard (inputs + Train + Check now + Reminder)
  • insights.tsx: Insights (trend chart)
  • settings.tsx: Settings (SecureStore PIN + time picker)
  • web.tsx: WebView screen

Data layer

• loom-expo/src/db/
  • client.ts: opens SQLite DB, creates schema
  • schema.ts: TypeScript row types
• loom-expo/src/repositories/
  • EventRepository.ts: inserts and queries events
• loom-expo/src/services/
  • RiskService.ts: computes risk + logs events
  • NotificationsService.ts: schedules a demo reminder
  • SecureStoreService.ts: demo PIN storage
  • analytics/AnalyticsService.ts: simple moving-average trend + anomaly note

---

Loom integration (on-device)

The Loom engine is integrated as a local React Native native module that bundles the prebuilt C-ABI binaries from 0_75/.

Where the native module lives

loom-expo/packages/loom-welvet/

Android

• Kotlin native module:
  • packages/loom-welvet/android/src/main/java/com/openfluke/loomwelvet/LoomWelvetModule.kt
• Bundled binary (arm64):
  • packages/loom-welvet/android/src/main/jniLibs/arm64-v8a/libwelvet.so
• The Kotlin module uses JNA to call into the C-ABI and exposes:
  • createRiskModel()
  • trainRiskModel(handle)
  • predictRisk(handle, features)

iOS

• ObjC++ native module:
  • packages/loom-welvet/ios/LoomWelvetModule.mm
• Vendored library + header:
  • packages/loom-welvet/ios/Welvet/libwelvet.a
  • packages/loom-welvet/ios/Welvet/welvet.h
• A podspec wires the vendored library/header:
  • packages/loom-welvet/loom-welvet.podspec

Important: Expo Go vs Dev Client

Because Loom is native code, you must use a dev client (or a built APK/IPA). Expo Go won’t include this module.

RiskService automatically falls back to a deterministic heuristic if NativeModules.LoomWelvet is not present.

---

How the “risk model” works (today)

This demo uses a very small Loom network:

• A single dense layer: 3 inputs → 1 output
• sigmoid activation → output is a probability-like risk score (0..1)
• It is trained on-device using LoomTrain with a small hard-coded dataset that reflects:
  • lower BG → higher risk
  • longer time since meds → higher risk
  • higher walking intensity → higher risk

This is deliberately small so it’s easy to validate the end-to-end integration (native module → JS service → UI → SQLite → chart).

---

Running the app (development)

From loom-expo/:

npm install
npx expo start

To actually use Loom on-device you must build a dev client.

Android (Windows)

cd loom-expo
npx expo prebuild -p android
npx expo run:android

iOS (macOS)

cd loom-expo
npx expo prebuild -p ios
cd ios && pod install && cd ..
npx expo run:ios

---

Building + installing a debug APK (Windows)

Two helper scripts exist at the repo root:

• build-apk-debug.bat: builds loom-expo/android/app/build/outputs/apk/debug/app-debug.apk
• install-apk-debug.bat: installs the APK to the currently connected adb device/emulator

Usage from repo root:

build-apk-debug.bat
install-apk-debug.bat

Notes:

• You need Android SDK + platform-tools (adb) on PATH.
• The APK produced is debug and intended for development/testing.

---

Building + installing a standalone Release APK (Windows)

If you want the app to run without Metro / expo start, build a Release APK:

• build-apk-release.bat: builds loom-expo\android\app\build\outputs\apk\release\app-release.apk
• install-apk-release.bat: installs it to the preferred adb device (auto-picks 192.168.* if present)

Usage from repo root:

build-apk-release.bat
install-apk-release.bat

Notes:

• This is still signed with the debug keystore (good for local testing, not Play Store).
• Release builds are slower than Debug, but the app is standalone.

---

Files added/changed at the repo root

• README.md (this file)
• build-apk-debug.bat
• install-apk-debug.bat
• build-apk-release.bat
• install-apk-release.bat