MooseTrax

Inspiration

For many, staying active is when we find ourselves most seen and grounded. Still, 43% of North Americans quit their fitness-related new year's resolutions by the end of January (that's coming up!). Specifically, accountability & safety measures like having a committed gym buddy, a paid personal trainer, or comprehensive fitness classes are often inaccessible, particularly as financial & geographic barriers to entry. We built MooseTrax because we ALL deserve accessible tools for keeping ourselves in tune with our physical identity.

What it does

MooseTrax is a personalized fitness coach that learns while you learn:

By uploading videos of your exercise routines, your Moose coach/cheerleader/data-analyst provides time-stamped annotations and recommendations for targeted strength training
The Dashboard stores your daily exercise streak, your physical strengths and areas of improvement, and recommendations for next exercises you should try based on your skills – as the user exercises more, MooseTrax builds a better profile on their strengths, weaknesses, and increases in difficulty for recommendations
Further, based on estimated skill level, the feedback tone shifts to use more accessible language to more beginner users

How we built it

Python (fastapi, mediapipe, cv2) for the backend video-processing algorithm
Gemini API for interpreting the dashboard database and improving based on behavioural analytics
HTML/CSS/JS for frontend development

Challenges we ran into

Determining an accurate and practical video-processing stack
Finalizing the scope and scaling methods of our web features

Accomplishments that we're proud of

The video processing successfully works by comparing the user to a data-scraped "ground truth" example of correct exercise form and passing a .JSON evaluation of position accuracy
Creating a website that ties together usability, character, and identity!

What we learned

Configuring LLM output, manipulating and storing databases, video processing

What's next for tbd

Further scaling to become more generalized! Potentially leveraging more training material and fine-tuning to provide structured analytics on different types of exercise disciplines (yoga, calisthenics, weight training, etc.)