Client
A North American venture-funded startup
Industry
Fitness & Wellness
Services
R&D, Software Development, Cloud/DevOps, Testing, Consulting
Tech
AOSP, Android, Kotlin, Java, Spring Framework/Boot/Cloud, Kafka, WebDriver/Selenium, Appium, Selenoid, React, Redux, WebSockets, Firebase, AWS, Jenkins, Docker, PostgreSQL, OpenCV, TensorFlow

Challenge

As life is getting insanely busier, making it difficult for many people to fit exercise into their routine, our client came up with an idea to build a futuristic fitness device for strength training at home. They wanted to create a smart fitness mirror that would be able to not only track exercise intensity or progress but offer users a “personal trainer inside,” powered by AI algorithms. The COVID-19 pandemic only amplified the urgency of their product as at-home fitness exploded after gym closures. After securing funding, the only thing the client needed was a reliable AI development company with extensive experience in deep learning and computer vision that could turn their idea into a successful product. And they found one — ITRex.

Our task was to:
Build the hardware components of the device and provide end-to-end software development services spanning infrastructure setup, embedded software/firmware development, and content management
Design and train a deep learning model using a dataset of workout records to provide feedback and guidance for users
Implement computer vision for motion tracking
Develop and build IoT sensors for weights to track time-under-tension, total load, and explosive strength
Develop an automation testing framework to provide comprehensive test coverage and accelerate time to market

Solution

A full-length wall fitness mirror with a custom AOSP-based operating system and custom drivers for the hardware is equipped with 3D cameras, a cutting-edge touchscreen, a voice-activated system, and an ML model. The mirror comes with IoT sensors and gives users access to live, on-demand classes and pre-programmed workouts.
Motion tracking and personalization
The front-facing cameras with ultra-wide angle lenses are powered by computer vision technology for 3D motion tracking to count the user’s reps and monitor form, with the system using depth sensing to capture motion based on anatomical landmarks
The proprietary ML model processes the landmark information to (i) learn the user’s capabilities for alternating between pushing them past their limits and encouraging rest when necessary, (ii) provide feedback on training sessions, and (iii) create tailored training plans for future workouts. The model uses a training dataset of hundreds of hours of workout records to make recommendations and adjusts them as it learns using the user’s historical training data
Sensors
Wireless adhesive sensors that users can stick onto their individual equipment to keep track of weights lifted and their total load
UI/UX
An LCD touch screen with a 4K UHD resolution provides an unparalleled user experience, ensuring crisp, clear visuals, while an in-built audio system delivers an exceptional audio quality, allowing users to play anything, from coaching audio to personal playlists
A voice control system. A voice-activated microphone can understand a fixed list of voice commands, effectively handling noisy environments. The speech recognition technology can learn new keywords to understand new commands given by the user
A custom companion iOS/Android application for users to share exercise results and receive feedback from other members of the community
Ability to work out with friends or a coach remotely via live streaming
Ability to integrate with smart watches and other fitness trackers to monitor the heart rate and other health data via the smart mirror in real time
Content management
A powerful admin panel, which allows moderators, coaches, and content providers to upload fitness content, assign workouts to users, and provide feedback on training sessions. The admin panel can be accessed from both a mobile device and a desktop
Testing
An automation testing framework that checks every functional aspect of the device, covering also the rep counting recognition technology and the ML model
AI-powered 3D motion tracking fitness mirror
Finess mirror

Impact

Acting like a personal fitness coach, this smart mirror is focused on giving serious training for professional results. The impressive breakthroughs achieved include:
Precise motion tracking
Real-time, exact rep counting
AI-powered feedback that provides both kudos and corrective suggestions
Adhesive sensor integration, allowing users to train with their own favorite equipment
Rapid generation of new, high-quality workout sessions
With these innovations, the mirror offers a modern, interactive fitness experience. It understands the user’s entire fitness journey—adjusting exercises when things get too easy or too tough, and intelligently planning future workouts to match progress and goals.

Latest projects