Rangkum AI 🤖📝

Rangkum is a powerful Android productivity tool designed to transcribe and summarize audio in real-time. It features a Floating Overlay that captures internal system audio (Zoom, Meet, Youtube, etc) and uses Offline Speech-to-Text combined with Generative AI to create concise summaries instantly.

It also supports YouTube Link Summarization via a custom Python backend.

---

📱 Screenshots

Home Screen	Floating Overlay	Summary	Tablet/Adaptive

---

✨ Key Features

• 🎧 Internal Audio Recording (Overlay):
  • Floating bubble interface (Picture-in-Picture style).
  • Captures system audio (Android 10+) directly without microphone noise.
  • Resizible & Collapsible window.
• 🗣️ Multi-Language Offline STT:
  • Dynamic Model Management: Support for 30+ languages (English, Chinese, Japanese, etc.).
  • On-Demand Downloads: Download only the languages you need to keep the app size small.
  • Smart Downloader: Features Resumable Downloads (handles network interruptions) and memory-efficient streaming using Ktor.
  • Privacy-focused: All audio processing happens on-device using Vosk.
• 🧠 AI Summarization:
  • Integrated with Google Gemini 2.5 Flash, Gemini 2.5 Flash Lite, Gemini 2.5 Pro (Paid).
  • Streaming Response: Text appears word-by-word (typewriter effect) for zero-latency feel.
  • Custom Markdown rendering for clean output.
• 📺 YouTube Summarizer:
  • Paste any YouTube link to get a summary.
  • Powered by a custom Python Backend (Flask + yt-dlp) hosted on Google Cloud Run.
  • Auto-detects video language & handles cookies for anti-blocking.
• 💾 Smart History:
  • Save chats locally using Room Database.
  • Swipe-to-delete with Undo (Snackbar) capability.
  • Search & Rename functionality.
• 🎨 Modern UI/UX:
  • Material 3 Design with Dark/Light theme support.
  • Adaptive Layout: Supports Foldables & Tablets using NavigableListDetailPaneScaffold.
  • Interactive Chat: Edit prompts, retry generation, and copy text blocks.

---

🛠️ Tech Stack

Android (Client)

• Language: Kotlin
• Architecture: Clean Architecture (MVVM + Repository Pattern)
• UI: Jetpack Compose, Material 3 Adaptive
• Dependency Injection: Koin
• Concurrency: Coroutines & Flow
• Network: Ktor Client (Backend API & Resumable File Downloader)
• Database: Room (SQLite)
• Local Storage: DataStore Preferences
• AI SDK: Firebase AI (Gemini)
• Audio: AudioRecord API, MediaProjection API
• STT: Vosk Android Library

Backend (Microservice)

• Language: Python 3.13.9
• Framework: Flask
• Tools: yt-dlp (YouTube Downloader), gunicorn
• Infrastructure: Google Cloud Run (Serverless)
• Storage: Google Cloud Storage (for cookie management)

---

🏗️ Architecture

The project follows strictly Clean Architecture principles to ensure scalability and testability.

com.achmadichzan.rangkum
├── data
│   └── # Implementation of Repositories, Data Sources (Room, Ktor, Gemini)
├── domain
│   └── # Business Logic (Use Cases, Repository Interfaces, Models)
├── presentation
│   └── # UI (Screens, ViewModels, Components, Service)
└── di
    └── # Dependency Injection Modules

Key Highlights:

• OverlayService: Handles the complexity of WindowManager, MediaProjection, and background audio processing.
• ModelRepository: Implements a robust Resumable Download logic using HTTP Range headers to handle large zip files and prevent OOM (Out of Memory) errors during extraction.
• AudioTranscriber: A dedicated class for managing the audio loop, Vosk inference, and thread synchronization (Mutex).

---

🚀 Getting Started

Prerequisites

1. Android Studio Otter or newer.
2. Device running Android 10 (API 29) or higher (Required for internal audio capture).
3. Firebase Project with AI enabled.

Installation

1. Clone the repository:

   git clone [https://github.com/achmadichzan/Rangkum.git](https://github.com/achmadichzan/Rangkum.git)

2. Setup Firebase:

   • Create a project in Firebase Console.
   • Download google-services.json.
   • Place it in the app/ directory.

3. Build & Run:

   • Sync Gradle.
   • Run on a physical device (Emulators often fail with Audio Capture).

---

🛡️ Permissions

The app requires sensitive permissions due to its nature:

• RECORD_AUDIO: To capture voice/system sound.
• FOREGROUND_SERVICE_MEDIA_PROJECTION: To keep the overlay alive and capture audio.
• SYSTEM_ALERT_WINDOW: To draw the overlay over other apps.
• INTERNET: For Gemini AI and YouTube Backend.

---

🤝 Contributing

Contributions are welcome! Please fork the repository and create a pull request for any bug fixes or feature additions.

---

Built by Achmad Ichzan & Gemini 3 Pro