Skip to content

Jeffuz/airsure

Repository files navigation

AirSure: AI-Powered Travel Assistant

AirSure is an advanced Android travel companion that leverages on-device AI to help travelers manage their airport experience. By combining real-time computer vision and local speech processing, AirSure provides automated carry-on size checking and proactive flight alert monitoring.

Team Members

Project Description

AirSure is designed to reduce the stress of air travel by providing intelligent, private, and offline-first assistance.

  • Vision AI: Uses YOLOv11 (sourced from Qualcomm AI Hub) to scan carry-on luggage against airline restrictions.
  • Audio AI: Employs a local Distil-Whisper model (sourced from Qualcomm AI Hub) to transcribe airport announcements and filter for specific flight alerts (gate changes, delays).
  • Edge Processing: The majority of the application logic and all AI inference runs locally on the device for maximum privacy and low latency.
Image

Key Features

1. Smart Carry-On Scanner (Vision AI)

Uses YOLOv11 via LiteRT to detect and analyze luggage in real-time.

  • Object Detection: Real-time identification of carry-on bags.
  • Restriction Management: Automatically compares detected items against airline-specific size regulations.
  • Hardware Acceleration: Optimized for Qualcomm NPUs via the Qualcomm AI Stack.

2. Live Flight Alerts (Audio AI)

Employs a local Distil-Whisper model to transcribe airport announcements.

  • On-Device Transcription: Privacy-focused audio processing using hardware-optimized models from Qualcomm AI Hub.
  • Pattern Matching: Intelligently listens for specific flight numbers for gate changes, delays, and final calls.
  • Live Transcription UI: Real-time visual feedback of the transcription process.

Setup Instructions (From Scratch)

Prerequisites

  • Android Studio Ladybug (2024.2.1) or newer.
  • Android SDK 31+.
  • Physical Device: Qualcomm Snapdragon 8 Gen 1 or newer recommended for NPU acceleration.
  • Git LFS: Ensure Git Large File Storage is installed to handle model files.

Dependencies

The project uses the following major dependencies:

  • LiteRT (TFLite): For on-device model inference.
  • Jetpack Compose: For the modern UI layer.
  • CameraX: For high-performance camera access.
  • Qualcomm AI Stack (QNN): For hardware-specific optimizations.

Build Steps

  1. Clone the Repository: git clone <repository-url>
  2. Pull Large Files: Run git lfs pull to ensure models are downloaded.
  3. Open in Android Studio: Let Gradle sync and download all dependencies.
  4. Configure JAVA_HOME: If building via CLI, ensure your path points to the bundled JDK in Android Studio (jbr).
  5. Run: Select your device and click Run (Shift + F10).

Run and Usage Instructions

Using the Carry-On Scanner

  1. Navigate to the Scanner tab.
  2. Point the camera at your luggage.
  3. The AI will detect the item and display its classification.
  4. View "Carry-on Status" to see if it matches standard airline dimensions.

Using Flight Alerts

  1. Enter your Flight Code in the Boarding Pass section.
  2. Navigate to the Alerts tab.
  3. The device will begin listening locally.
  4. When an announcement is detected, it will be transcribed live.
  5. If a match is found for your flight, an alert card (Gate Change/Delay) will appear.

Technical Stack

  • Language: Kotlin & Java
  • UI Framework: Jetpack Compose (Material 3)
  • AI Engine: Google LiteRT
  • Models: YOLOv11 (Vision), Distil-Whisper (Audio) - Both optimized via Qualcomm AI Hub.

Tests and Verification

  • Unit Tests: Located in app/src/test. Run via ./gradlew test.
  • Logic Verification: Use the Debug Menu in the Flight Alerts screen (click the "Listening" indicator) to simulate test announcements via pre-loaded assets.

Notes

  • Privacy: All audio recording and transcription stays on-device. No data is sent to external servers.
  • Performance: NPU acceleration is automatically selected if supported by the hardware, falling back to GPU or CPU otherwise.

References

License

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

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors