Recently I was trying to step up my YouTube game with cool new 3d graphics and visual effects. I thought: “Wouldn’t it be cool if I just wrote a program that could just read my hand gestures from the screen and render in 3d the AWS Architecture I was referring to”. Then I realized I was doing what I always do… over engineering the heck out of it.

But before I decided not to fall into my old overengineering habits I found this cool model that runs in the browser. It is a Hand Pose Estimation model that can read your webcam and return an array of x,y coordinates for each of your hands joints.

I send this out, not so much to give you a tool to draw green dots on your hands, but to try and get you thinking about what else we could build classifiers for. Think about niche situations for your clients.

What about an LLM that could score in a gymnastics competition? One that knew what the perfect Takachive looked like. Or what to score a Pak Salto if their knees were bent.

Just some food for thought.