Posture Reminder (aka PostureNet): our story

Who is behind PostureNet?

Hi, my name is Leon Wei, I built Posture Reminder (aka PostureNet) to fix my own posture problem.

Most recently, I was a senior manager of applied machine learning at  Apple, where I lead teams of data scientists, software engineers and product managers building large-scale machine learning systems for Apple's billion dollar businesses.

I also worked as a research scientist at Amazon, where I developed its real-time pricing engine for millions of products sold there.

Please feel free to connect with me on Twitter or LinkedIn

Why did you create PostureNet?

I was sitting 10+ hours a day in front of computers while working at  Apple (left early 2021).

During those long work hours, I often focused on the work too much that I forgot about sitting straight.

As a result, I was constantly in a slouching posture, leading to severe back pains and neck issues over the last few years.

One day my chiropractor told me that I need to be focusing on good posture, so I told all of my team members to watch out for my posture and yell at me if they see me slouching.

It worked well. Being publicly shamed by slouching has kept me pay a lot of attention to my posture.

My back was straight, my neck no longer feels awkward, and my eyes no longer dry.

All good until Covid-19 hit.

I fall back to bad postures because no coworkers around to watch it for me, and I need something to help me out.

That's when I started working on PostureNet, an AI based on the latest computer vision and machine learning techniques that will monitor my own posture in real-time.

Since I started using PostureNet to help fix my slouching posture, I was able to keep a good posture during the day, and those neck and shoulder pain gradually went away.

I hope you give PostureNet a try, and please feel free to reach out to me on Twitter, if you have any questions or feedback.

Technically, how does PostureNet work?

Great question.

PostureNet is a deep neural network based on the latest computer vision and AI technologies.

There are two critical phases to create a good performing PostureNet, a machine learning 'model'.

1. Configuration phase (training in machine learning terminology) when you first use the app, you are asked to share your good and bad posture data with PostureNet.

It then tries to learn what specific features to best differentiate your healthy posture from slouching.

For example: when you slouch, your shoulders are more rounded, PostureNet analyzes and converts many of those 'signals' into numerical values and keeps a copy of them on your device.

2. When you activate Posture Net, it moves on to the prediction phase (Inferencing in machine learning terminology).

It tries to associate your real-time pose with a good posture or a bad one with a probability (a score you see in the app), and when the score reaches 90%, it almost certain that you are in a bad posture and reminds you.

If you have other technical questions, please feel free to reach out.