A High Schooler’s AI-Infused Running App: Fast and Fit
December 21, 2022
Joshua Joseph is like a lot of you. Smart and super passionate about running, he loves the LetsRun.com forums. He’s always trying to figure out ways to improve his running.
Where Joshua, a 17-year-old senior at Foothill High School in Pleasanton, California, stands out is he has developed an app called Fast and Fit that helps runners improve in two distinct ways.
First, the simple way Fast and Fit can help you- it has a race calculator that converts times from 600m up to the 10k, using Pete Riegel’s formula. Handy and helpful for sure.
Now for the more advanced. Fast and Fit also uses AI (machine learning) to analyze the quality of your running shoes. You take a picture of your shoes via the app and it can identify the quality of the shoe and any problems like overpronation or oversupination based on the pattern of the wear of the shoe. As Joshua says, “It is the first app that analyzes running shoes using machine learning and is useful for runners of all levels.”
Considering LetsRun.com has been around 22 years and still doesn’t have an app, we decided to learn a little more about the high schooler who designed Fast and Fit.
The app is currently available for iphones here.
LetsRun.com (LRC): First, congratulations on the app. What prompted you to make it?
Joshua: I am an avid runner and I have been running cross country and track for the past six years. I have been a varsity runner for the past four years. As a runner, I had some injuries which affected my race performances at meets. I did some research online and found that the probable cause of many of my injuries and poor performances were due to running form and worn out running shoes. I learned that running form problems like overpronation and oversupination caused injuries and wore out running shoes faster. These problems can be identified by the wear pattern on a running shoe. I wanted to help other runners dealing with injuries to fix their running form and prevent common injuries like shin splints or plantar fasciitis. As I learned programming in AP Computer Science at school and iOS app development through an online course, I realized I could use computer science to solve this problem. This led me to develop this app as a way to help runners spot and deal with these injuries faster.
LRC: As for the app itself, it seems to do two very different things. You have the race predictor calculator and then the shoe wear and tear analyzer. Was the race calculator easier to make?
The race calculator was easier to make than the shoe feature. For the race calculator I applied Pete Riegel’s race prediction formula to predict times from one race distance to the other. The formula was useful because it can be used to convert times between any distance. The shoe feature was more challenging to create because it took a lot of time to train & improve the accuracy of the AI model.
LRC: The race calculator stops at 10k. Clearly, you’re not yet at the age where everyone becomes obsessed with their marathon times. Any plans to expand it to the marathon and half marathon?
Definitely. I developed the app with cross country and track runners in mind so I set it up for those distances. I plan to expand the race calculator to half marathon and marathon distances in the future.
LRC: As for the shoe wear and tear, your app processes images and has AI. How the hell did you learn that and how does it work?
As I began to learn iOS (Apple) programming, I started to explore the use of AI models available in iOS — CoreML (Core Machine Learning). My interest in these AI features grew as I learned about key CoreML features like Object Detection, Image Classification, etc. that are used in the detection and recognition of objects. I developed the shoe feature by training the Image Classifier model. This was done by collecting many images of soles of running shoes in various states of wear and tear. The images were classified into four groups:
- “worn out from overpronation”
- “worn from oversupination”
- “hole or tear”
- “good shoes”
After training the Image Classifier model using these categories, I improved the accuracy of the shoe feature by testing the feature on my teammates’ shoes.
LRC: In terms of your own running, what events do you do? How fast are you? What do your teammates think of the app?
In track & field, I run the mile and 800m and my best times are 2:10 for the 800m and 4:55 for the mile. In cross country I run the 2 mile, 3 mile, and 5k and my best times are 16:52 for 2.93 miles and 17:24 for 3 miles. My teammates tried the shoe feature and were able to identify problems with their running shoes. They found the app useful. Some of my teammates suggested a few improvements which I have incorporated in the app.
LRC: What are your plans for college?
I plan to major in computer science and continue running in college.
LRC: Good luck and stay in touch.
Joshua’s app was selected as a finalist for California District 15 in the Congressional App Challenge.