AthleteAI | Devpost

Inspiration

As avid athletes, specifically tennis players, we wondered if there was a way to improve our game and model it after professionals without having to worry about the hassle that comes with personalized coaching. We were inspired to create this after seeing the endless possibilities of new innovative AI tools like DeepMotion and LanceDB.

What it does

AthleteAI allows users to have a side-by-side comparison of their swings compared to different professionals. We have preinstalled training models for three prominent tennis swings: service, forehand, and backhand. Users can upload their own video clips and view how their swings differ from the professionals. Further, if users need help to improve their swing or model it after the professional, they can chat with our trained AI chatbot model that can give them personalized feedback based on their description of the problem.

How we built it

We used HTML and CSS for the overall website design. The backend work was handled using JavaScript and Python. DeepMotion’s API allowed us to visually display the 3D model of the swing on the website. LanceDB’s API allowed us to train the chatbot that gives personalized feedback.

Challenges we ran into

The main challenge we ran into was the constant back and forth between different visual formats for the video of the swing. Going from mp4 to inputting it into the API to using the resulting fbx file and displaying it on our website was the biggest challenge. We were able to overcome this using different packages and different file conversion methods.

Accomplishments that we're proud of

We are most proud of creating a product that we envision ourselves using and hopefully something anyone passionate about tennis, or any other sport, can use. We are also incredibly proud of the group effort that went into creating this project.

What we learned

We learned how to effectively navigate going through extensive documentation when trying to understand how to use a product. This was most of our first time participating in a hackathon. We realized the extremely difficult but eventually useful learning curve of understanding other developers’ documentation and implementing it in our projects.

What's next for AthleteAI

There are a few things we hope to implement next. The first is being able to visually overlay the user-uploaded video and the training video over each other so users can easily identify differences in their swings instead of having to estimate different angles and other differences between swings. The other future implementation we hope to add is expanding to other sports. We see this as a valuable tool for many sports like golf, table tennis, swimming, and any other sport where physical form makes a huge difference, and learning from professionals can exponentially grow your game.