Act.App - AI powered app for Healthcare


An AI-powered application to monitor and analyze wearable and smartphone sensor data using self-supervised algorithms to evaluate and enhance the activity level of a person in a personalized manner. In particular, the project is divided into two parts: the first part, a Proof of Concept of the application that through gamification and education guides the users to improve their well-being and healthcare. The second part is the improvement of the existing algorithms for human activity recognition using Self-supervise learning methods.

Technologies used

Tool for the app: Plots on the app: Python and Matplotlib Tools for algorithm development: Python, Github and Google Colab AI Frameworks: TensorFlow and PyTorch. Algorithm Specifications: Self Supervised Learning, GANs and SelfHAR.


Time - not only the restriction of one weekend, but also the different time zones of the team members; Team Management - Divide tasks, work with people with different backgrounds, communication during the development.; Understanding the state of the art work of the algorithm and how to improve it.


Clear app prototype with the pizza animation and the drawings of the people's avatars on the social part, which uses the activity classification to engaje the user in a fun way. We managed to work together at the end and achieve the goal we set for the weekend; Add improvements to state of the art achievements results for the human activity recognition algorithms.


For the algorithm, we understood better how the self supervised learning works for signals and how different techniques, like transfer learning, could be used to improve model performance where there are no big labeled datasets available. We learned to work together as a team. We also learned how to better engage the user to improve their health. And how to manage colors better on the app to showcase the need for less sedentary behaviour and more activity during the day. We learned more about the SGDs.

Next steps

Implement the app prototype and integrate it with the developed algorithm for the activity classification. Make the algorithm and the app open, so everybody can integrate with new ideas and other best practices.