The tech stack | HaptiBraille

Languages

Programming languages and what they are used for
Language	What it does in HaptiBraille
Dart	Almost everything: every screen, the Braille translator, the timing engine, training, analytics, storage and the tests.
Kotlin	The Android code that talks directly to the vibration motor (`HapticChannel.kt`).
Swift	The iPhone app shell, generated by Flutter. The iPhone vibration code is still to come.
Gradle Kotlin DSL	Android build settings: app ID, Java version and SDK versions.
XML	The Android manifest: app name and permissions for vibration and the microphone.
YAML	The package list, font list and versions (`pubspec.yaml`).

Framework and tools

Frameworks and tools
Tool	Version	Why
Flutter	3.47 stable	Google's toolkit for building one app for Android and iPhone from a single codebase.
Dart SDK	3.13	The language Flutter apps are written in.
Android SDK	API 36	Builds and runs the Android app. Minimum supported phone: Android 7 (API 24).
Android NDK	28.2	Native build tools required by Flutter's Android build.
Gradle	9.3	Builds the installable APK file.
Java (JDK)	17	Runs Gradle and the Kotlin compiler.

Packages

Dart packages used
Package	What it is	What we use it for
flutter_riverpod	State management	Shares the timing settings, playback, training session and progress between screens, and lets tests swap in fakes.
hive, hive_flutter	On-device database	Saves timing settings and every training answer as plain JSON.
speech_to_text	Phone speech recognition	Talk to me, dictation, and spoken training answers.
fl_chart	Charts	Accuracy and reading-speed lines, and the per-letter accuracy bars.
clock	A replaceable clock	Times playback and answers in a way tests can control.
flutter_test, flutter_lints	Testing and code style	28 automated tests, and zero analyzer warnings.

We deliberately didn't use a ready-made vibration package. Those usually offer "vibrate for X milliseconds". We need exact rhythms with different strengths played as one uninterrupted pattern, so we wrote a small bridge straight to Android's own API.

How the code is organised

The app follows Clean Architecture: separate layers with one job each. The core rules (what Braille is, how timing works, how answers are scored) don't depend on screens, storage or hardware, so they are easy to test and easy to swap.

ScreensWelcome, Home, Read, Talk to me, Train, Progress, Settings, and the animated Braille cell.

ControllersRiverpod providers that run playback and training sessions and hold your settings.

Domain (the brain)Pure Dart: Braille translator, timing engine, curriculum, analytics, data models.

DataHive storage, speech recognition, the vibration bridge, the AI agent placeholder.

AndroidKotlin HapticChannel plays the waveform on the motor.

The heart of it, in code

This Dart function turns one Braille cell into a rhythm. It is the whole idea of temporal Braille in about twenty lines (lib/domain/engine/haptic_translation_engine.dart):

if (p.cellAnchorEnabled) {                 // soft start-of-letter tick
  out.pulse(anchor(p.anchorDuration, p.anchorIntensity));
  out.silence(intra);
}
for (var slot = 0; slot < 6; slot++) {           // the six dots, in order
  if (slot > 0) out.silence(intra);              // gap between dots
  if (cell.isRaisedAt(slot)) {
    out.pulse(dot(p.dotDuration, p.intensity, slot)); // raised: tap
  } else {
    out.silence(p.emptySlotDuration);                // flat: silence
  }
}

And this Kotlin code plays the result on the phone (android/…/HapticChannel.kt). On Android 13 and newer it is marked as accessibility vibration, so it still works when the user has turned off touch vibration:

val effect = VibrationEffect.createWaveform(timings, amplitudes, -1)  // -1: play once
if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.TIRAMISU) {
    v.vibrate(effect, VibrationAttributes.createForUsage(USAGE_ACCESSIBILITY))
} else {
    v.vibrate(effect)
}

Ready for AI

All text passes through one interface, ITextProcessor, so an AI could summarise a long email before it is vibrated.
Learning data exports as versioned JSON that an AI can read to spot where someone struggles.
HapticProfileModel.scaled() slows or speeds up every pause at once, so an AI coach could adjust the reading speed automatically.
An empty AgentService marks where the AI connection will go.

This website

Plain HTML, CSS and JavaScript with no framework and no build step. The background dot field is drawn on a canvas. The "Feel it" demo uses the same timing rules as the app and the browser's Vibration API, which works in Chrome on Android. Fonts: Bricolage Grotesque for headings and Atkinson Hyperlegible for text.

All the source code for the app and this site is on GitHub.