X-Shred - Voice Assistant
X-Shred Mobile Voice Assistant Designed and developed an offline-first mobile voice assistant integrated into the X-Shred application to enhance the on-mountain skiing experience.
Impact
10,000+
Condidates Screened Monthly usding AI
$400,000+
Annual savings on recruitment costs
30%
Improvement in EBITDA margins
The system enables users to interact naturally with the app through voice commands, even in low-connectivity environments common in mountain regions.
The assistant processes queries directly on-device, allowing users to retrieve contextual information such as current location (“Where am I?”), navigation guidance (“How do I get to [lift/run]?”), and operational details (“How long is this lift open?”). The solution prioritized low-latency responses, reliability without network dependence, and a seamless user interaction model.
This project combined mobile development, on-device inference, and voice interface design to deliver a responsive and practical real-world assistant.
Impact
Achieved sub-second inference latency on mobile devices
Maintained high recognition reliability in noisy outdoor environments
Reduced false wake events compared to Porcupine
Improved wake word detection accuracy
Enabled fully offline voice interaction & navigation workflows
Delivered low-latency, on-device voice processing
The Challenge
Ski environments present unique challenges for voice-enabled systems.
Connectivity is often unreliable, and environmental noise can significantly impact speech recognition. Key challenges included:
Limited or no internet connectivity on mountains
High background noise (wind, movement, crowds)
Need for real-time, low-latency responses
Difficulty integrating voice interaction seamlessly into a mobile app
High false wake rates in existing solutions
The goal was to create a reliable, offline-first voice assistant that works naturally in extreme outdoor conditions.
Approch
We designed an on-device AI system optimized for low-latency and offline performance.
Key approach elements:
Implementing on-device inference for speech recognition and intent handling
Integrating a custom wake word detection system (optimized beyond Porcupine)
Designing a voice interaction layer tailored to skiing use cases
Optimizing models for noisy, real-world environments
Seamlessly integrating with the X-Shred mobile app
The objective was to ensure fast, reliable, and intuitive voice interaction without network dependency.
Solution
We developed a fully offline mobile voice assistant with:
On-Device Voice Processing
Handles speech recognition and intent understanding locallyContext-Aware Responses
Answers queries like location, navigation, and lift informationNavigation Assistance
Guides users to specific lifts, runs, or locationsCustom Wake Word System
Improves activation accuracy and reduces false triggersLow-Latency Interaction Engine
Ensures near-instant responses for a smooth user experienceSeamless App Integration
Fully embedded within the X-Shred ecosystem
Result
Delivered a production-ready offline voice assistant
Enabled reliable voice interaction in extreme outdoor conditions
Improved usability and accessibility of the mobile app
Enhanced overall skiing experience through hands-free interaction
Let’s build something that matters with speed and clarity
Tell us what you’re working on and we’ll explore how
our team can help bring it to life with AI and UX