AInSight - AI-Powered Mobile Assistant

AInSight is a privacy-first, on-device AI assistant for Android that combines text chat, voice transcription, and team collaboration features using the RunAnywhere SDK.

Features

🤖 AI Assistant

• On-device LLM inference using RunAnywhere SDK
• Real-time streaming responses for natural conversations
• Multiple model support with automatic CPU optimization
• Privacy-first - all processing happens locally on your device

🎙️ Voice Notes

• Voice-to-text transcription with RunAnywhere's speech recognition
• Task extraction from voice recordings using AI
• Permission-based recording with intuitive UI
• History management for all voice notes

👥 Team Chat (Coming Soon)

• Collaborative AI conversations with team members
• Shared voice notes and task management
• Team model preferences and settings
• Real-time messaging integration

Technical Stack

• Language: Kotlin
• UI Framework: Jetpack Compose with Material 3
• AI SDK: RunAnywhere SDK v0.1.3-alpha
• Architecture: MVVM with StateFlow
• Navigation: Jetpack Navigation Compose
• Permissions: Accompanist Permissions

Setup Instructions

Prerequisites

• Android Studio (latest version)
• JDK 17+
• Minimum Android SDK 24 (Android 7.0)
• Device with at least 2GB RAM (recommended 4GB+ for larger models)

Installation

1. Clone the repository

   git clone <repository-url>
   cd ainsight

2. Open in Android Studio

   • Open Android Studio
   • Select "Open an existing project"
   • Navigate to the project directory

3. Sync the project

   • Android Studio will automatically prompt to sync
   • Click "Sync Now" to download dependencies
   • Wait for the RunAnywhere SDK to be fetched (first build takes 2-3 minutes)

4. Run the app

   • Connect an Android device or start an emulator
   • Click "Run" in Android Studio

First Launch

1. Grant permissions when prompted (microphone for voice features)
2. Download a model - the app will register small models automatically:
   • SmolLM2 360M (119 MB) - Fast, basic responses
   • Qwen 2.5 0.5B (374 MB) - Better quality conversations
3. Start chatting with your on-device AI assistant!

Project Structure

app/src/main/java/com/example/ainsight/
├── MainActivity.kt              # Main activity with navigation
├── MainApplication.kt           # Application class with SDK initialization
├── AssistantScreen.kt           # LLM chat interface
├── AssistantViewModel.kt        # Chat logic and state management
├── VoiceNoteScreen.kt          # Voice recording interface
├── VoiceNoteViewModel.kt       # Voice processing logic
├── TeamChatScreen.kt           # Team collaboration (placeholder)
└── ui/theme/                   # Material 3 theming
    ├── Color.kt
    ├── Theme.kt
    └── Type.kt

SDK Integration

RunAnywhere SDK Setup

The app uses the RunAnywhere SDK for on-device AI capabilities:

// SDK Initialization in MainApplication.kt
RunAnywhere.initialize(
    context = this,
    apiKey = "dev",
    environment = SDKEnvironment.DEVELOPMENT
)

// Register LLM provider
LlamaCppServiceProvider.register()

// Register models
addModelFromURL(
    url = "https://huggingface.co/...",
    name = "Model Name",
    type = "LLM"
)

Available Models

• SmolLM2 360M Q8_0 (119 MB) - Quick testing and basic Q&A
• Qwen 2.5 0.5B Instruct Q6_K (374 MB) - Balanced performance and quality
• Llama 3.2 1B Instruct Q6_K (815 MB) - Higher quality conversations

Key Features Implementation

Streaming Chat

// Real-time response streaming
RunAnywhere.generateStream(prompt).collect { token ->
    assistantResponse += token
    // Update UI in real-time
}

Voice Permissions

// Permission handling with Accompanist
val audioPermissionState = rememberPermissionState(
    android.Manifest.permission.RECORD_AUDIO
)

Navigation

// Bottom navigation with 3 tabs
sealed class Screen(val route: String, val title: String, val icon: ImageVector)

Development

Building the Project

./gradlew assembleDebug

Running Tests

./gradlew test

Code Style

The project follows standard Kotlin coding conventions with:

• MVVM architecture pattern
• StateFlow for reactive state management
• Jetpack Compose for declarative UI

Performance Considerations

Model Selection

• SmolLM2 360M: Fastest inference, suitable for basic tasks
• Qwen 2.5 0.5B: Good balance of speed and quality
• Larger models: Better quality but require more RAM and processing time

Device Requirements

• Minimum: Android 7.0, 2GB RAM
• Recommended: Android 10+, 4GB+ RAM
• Optimal: Android 12+, 6GB+ RAM with dynamic colors

Privacy & Security

• On-device processing: All AI inference happens locally
• No data transmission: Conversations never leave your device
• Local model storage: Models are downloaded and cached locally
• Permission-based: Only requests necessary permissions

Troubleshooting

Common Issues

1. Model download fails

   • Check internet connection
   • Verify sufficient storage space
   • Try a different model

2. App crashes during inference

   • Close other apps to free memory
   • Try a smaller model (SmolLM2 360M)
   • Restart the device

3. Voice recording not working

   • Grant microphone permission in Settings
   • Check device microphone functionality
   • Restart the app

Performance Optimization

• Enable "Large Heap" in manifest (already configured)
• Close background apps during AI inference
• Use Wi-Fi for model downloads to avoid cellular data charges

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• RunAnywhere AI for the on-device AI SDK
• Google for Jetpack Compose and Material Design
• Hugging Face for hosting the AI models
• Open source community for various libraries and tools

Support

For support, please:

1. Check the troubleshooting section above
2. Review RunAnywhere SDK documentation
3. Open an issue in this repository
4. Contact the development team

---

Built with ❤️ for privacy-conscious AI users