What happens when three wine lovers attend the same professional development lectures? They come up with a new idea on how to revolutionize wine tasting!

Did you know that if you tasted a single wine a day, you would only scratch the surface of the iceberg of brands available? Magazines with articles about all kinds of wines have been around for a while helping us make the decision on which bottle to buy. The internet revolution brought us blogs and public reviews on a more fine-grained level. The name Gary Vaynerchuk is to many no longer unknown, as he built one of the first successful online e-commerce platforms selling wines. And he made himself a name with his youtube videos where he carefully explained how to taste wine and which wines he likes.

Now, the smartphone revolution and rise of (careful, buzzword approaching) AI opens up wine tasting to a whole new world. How about a personal companion who learns from your wine ratings and helps you pick the one which fits best to you?
This sounds like a pretty big vision but this team of entrepreneurs feels up to tackle the challenge.

Approach

Finding the right people for tackling this problem was not that easy. Our usual target group consists of students out of which only a handful are devoted wine lovers.
After the Design Thinking Workshop, we narrowed down three main areas of interest which are all part of the user journey.

First, we start with the onboarding (business jargon for the registration process) of new users. Then we have the usage of the app for recognizing a bottle of wine followed by the recommendation system. All three areas are very important. Apps for tracking wine consumption and giving public ratings already existed but their focus lied only on the recognition. If you can get a basic taste profile of the user throughout the registration process you gain an immense advantage to bring the best recommendations.

Results

The requirements of a good and flexible solution made it clear that the companion is an app on your smartphone. All the magic and number crunching happens in the shadows of a background server.

Most wine recognition and rating apps only work when taking a very close up picture of a wine bottle. Going through a selection of wines is done in a one-by-one manner which can become very cumbersome. Just imagine you’re standing in front of a wine shelf with a dozen brands. This means that there is still some room for improvement. We kept the neural networks for the recommendation systems and implemented a bottle recognizer using some classic feature matching techniques using OpenCV.

The set of images of existing wine bottles was crawled from online shops. After some tweaking, the solution worked quite well and even recognized multiple bottles in a single image correctly.

For the recommendation system, we got inspired by approaches of leading players in this field such as Netflix and Spotify. Users with similar profiles will get similar wines they didn’t try out yet recommended.

We wish the digital vino team all the best on their further journey. May they revolutionize wine tasting and help everyone with picking the right wine.

‍