Building a useful personalized recommendation system is not an easy task. There're many factors to consider to pick the right approach and make it work in particular business conditions. But in simple words the paradigm is the same: if the users with similar preferences as I like some product, there's a high probability that I will like it too. The more data about your products and your users you have, the more precise the recommendations will be.
We've made a basic demo of a recommendation system to show how it works. It is easier to simulate and understand by example of music recommendations. As a source of data we used an open source dataset from 2012 by Last.fm service. For the development we used Python, Streamlit, Numpy, Pandas and SQLike.
The similar approach can be used for any other industry. The idea behind the ML algorithm in the demo is to detect the dependencies of users preferences. If you type an artist name the demo will recommend similar artists often listened to by people who are also listening to this artist. Just enter your name and e-mail and check yourself.