AI Powered Personal Coach Application

Time and Materials project based engagement. A custom solution made as a side project of Gym network in UAE. We were lucky to get an amazing experience with great video materials to train an ML model.

Computer Vision Application for Personal Training

Services provided by Celadon: machine learning, computer vision, technical and graphic design, project management, frontend development, backend development, QA.

Gym Business Revolutionizing Concept

Features, media data processing, 2D points of main exercise position transforming into a 3D, video capturing and comparison, analytics model performance optimization, progress tracking and calendar exercise planning.


1 Frontend Engineer
1 Backend Engineer
1 Machine Learning Engineer
1 QA Engineer/ PM


4 months


UAE, Dubai

Tech Stack

C++, Python, TensorFlow, TensorflowLite, React.js, React Native


There are thousands of people all over the world who want to work out properly but simply don’t know-how. They cannot afford a personal instructor or may not keep an eye on all the intricacies of performing an exercise. As a result, they do not achieve set goals and may even give up on sports due to the disappointing experience in the past.

With this in mind, our client had a unique solution – an AI-powered virtual assistant. He approached Celadon with this idea and we immediately thought we could develop an AI-powered mobile app. With the help of artificial intelligence, the app would closely monitor the user via the camera and would compare the movements of the user to the ones we would upload as references – the movements of professional athletes and gym coaches. 

In addition, the app would also be able to monitor the user's progress and tailor an individual workout program just for the user and for all those who wish to work out in a fun and efficient way.

Example: a lady performs a set of exercises aimed at abs. The app will display the match of the user with the instructor and provide useful recommendations on improvement.


  • To create a custom dataset so that the app is able to monitor and record certain movements and compare them to the ones that are indicated as references
  • To develop an optimized architecture
  • To translate a 2D body point into a 3D one by a separate model
  • To compare the instructor’s video and the received data from the user’s device


By using a set of tools listed above, we built a platform that processes the video data of the exercises that are performed by an instructor and also translates the 2D body point functions into 3D. 

Our app features a catalog of exercise templates for the user and can capture the video of the user performing the movements. 

The movements’ analysis and comparison happen in the server and the processes video is then returned to the user with personalized recommendations.

Drop us a message

and we will get back to you
within next 12 hours

Enter name
Enter phone
Enter email
Enter message
attach file
What do people say?
They read our minds and provide exactly what we want. The ease of the process is wonderful....