Tuffr | Devpost

Inspiration

We set out to make scientific training accessible to everyone. No one should ever feel embarrassed or out of place in a gym; everyone deserves the chance to learn, grow, and train with confidence. To solve this, we created Tuffr to help people analyze their lifts and make sure they can train without injury.

What it does

Tuffr lets you choose a workout and sends that information to our AI agent. Using a computer vision model, it tracks your movements and gives you real-time feedback on your form, along with suggestions for improvement. On the left side, you’ll also find a chatbot you can ask for guidance, whether you want to know what to do next or get tips to improve your lift.

How we built it

We built Tuffr with a React frontend enhanced by smooth animations using Lenis. Our backend is powered by a Python-based AI agent and chatbot. For computer vision, we use OpenCV with YOLO to detect key points and calculate angles that represent the user’s form. This information is then sent to the frontend, where we display the user’s form score in real time.

Challenges we ran into

We've had various problems with the ECG electrode configuration and the other software bugs, and we overcame these issues very quickly and were able to create a cohesive product that was able to successfully measure and give data on a person's workout. To go more in-depth, the ECG pads output a very low-strength voltage signal, which we had to amplify using op amps to get appropriate readings.

Accomplishments that we're proud of

Overall our team successfully built the project and was able to give analytics on how the workout went. We increased accuracy for the computer vision model.