SKITAG AI

Your DATA. Your ASSET. Your AI.

Skitag in short

It started back in 2016/2017 as a project to collect input datasets in streaming and use it to train Deep Learning algorithms. First, we fixed an Arduino IMU on a ski board, and connected it to an iOS App. We started to see a huge expectation from skiers regarding the idea of having information in real time about their carving. We designed our own IMU device and started to think about Skitag as a product. Then, we implemented AI on the device, both inference and learning (i.e. backpropagation). We extended this experience to Surf and Bike. Unfortunately, the IMU device didn’t work as expected (i.e. issues regarding the fixing system and IoT configurations). So we started to ask: What if the smartphone could also do the job of collecting, training, and visualizing the data? In 2023 we designed an iOS App that can do all the jobs (i.e. collect, process, train, visualize and update). We introduced the counterintuitive concept of having millions of customized AI engines, instead of having one AI engine trained with millions of different input datasets. We hope that you enjoy this Augmented Ski Experience as much as we enjoy doing this job.

Mike

SKITAG, SURFTAG, ... throughout the years

2024

Share your CARBIO. Create a QR pass code to share your CARBIO or your DOWNHILLS to your friends and family. Let they scan your QR code. That's it!

2023

Carving Biometrics. Multi-device (iOS, iPadOS). Customizable In-App AI Engines (from scratch). AI Ethics. Your DATA. Your ASSET. Your AI.

2021-2022

Bits rather than atoms! Use the mobile device as a sensor.

2020

Build a brand-new IoT sensor. Implement on-device training. Predictive Carving. Skitag & Surftag Early Adopter Program

2019

Gather data. User Experience oriented. Ski, Bike and Surf.

2018

First Skitag's prototype. First App with AI inside (iOS + Android). Skitag 2018 Augmented Ski Experience

2017

The Inception. iOS + Arduino.

The Whole Story

It started back in 2016/2017. What was the problem to be solved? Back then it was very difficult to find data in streaming to test Machine Learning (ML) Engines. In this case, input datasets to train both Deep Neural Networks and Long Short Term Memory (LSTM) algorithms. We thought that the information collected from an accelerometer and a gyroscope could be a very good source of information to start with. It certainly did the job. We managed to test our own ML libraries built in Java + Apache Spark, and we found some very interesting metrics.

At first, it was kind of a hobby. But then, we found that every skier we presented the idea to was excited about the idea of having information on real time about their carving, their speed and their efficiency while they were going downhills. It was like a WOW moment. We’ve got something interesting. So we started to think about the idea of making it a product to sell. We started with the design of an Inertial Movement Unit (IMU) to fix into the ski board. There were many tries. Many devices were lost. The idea was good, but there was something not going as expected.

2021 and 2022 was an impasse period. Even though we tried a few more prototypes of the IMU device, we started to think: What if the smartphone could do the IMU’s job? So we designed a new iOS App to collect and visualize the data. On the other hand, we already had our own ML Engines libraries developed in Swift language. So we thought: What if the smartphone could also do the job of collecting, training, and visualizing the data? We already had a backpropagation for an LSTM working in Skitag’s App of 2020. Why not complete the whole pipeline? Machine Learning Operations (MLOPs) in your smartphone.

Yes, we did it. 2023 Skitag’s App has a complete MLOPs implementation. You collect your own data. You process your input datasets (i.e. labeling). You train your own AI Engine. You test your ML Engine (your CARBIO). You can update your CARBIO, remove it and start back again. Your DATA. Your ASSET. Your AI. All in your own device. Sounds good, but it could also sound counterintuitive. Sounds like overfitting?

Could be. Those who work in the AI field are very aware of the nightmares derived from the difficulties to generalize well to every skier in the world. Not to mention that every skier has its own style, its own skis, they carve in different ways and not always do it under the same snow conditions. That was something that we heard from them (BTW: thanks a lot for sharing that). So overfitting may not sound that bad in this case. As a matter of fact, it could solve the problem of generalization in this field.

Right, Skitag 2023 works counterintuitively. It overfits each of the skiers using the application. There is not only one ML engine doing the job of finding patterns in millions of skiers. Instead, within the Skitag ecosystem there are millions of ML engines finding customized patterns. It’s like an extreme federated system where each device is training an algorithm, but instead of sharing that information to build one ML engine capable of generalizing to all the skiers, the information is stored and used locally. That means that the customized AI engines are lighter, easier to train, and therefore more efficient than training and updating one ML engine that generalizes well to the whole world.

What's next? We are planning to let you have more than one ML Engine. So you can have a CARBIO for some special conditions, skis, snow, and whatever you may think could be appropriate. Moreover, we are thinking about the possibility of building deeper AI engines to better predict your movements. But first we would like to know if you are comfortable with this Augmented Ski Experience. Please, share your thoughts in our Instagram profile. We'll be watching.

We hope that you enjoy this Augmented Ski Experience as much as we enjoy doing this job. Take care.

Mike [in]

PS: Skitag means "ski day" in German.