The underlying idea is quite straightforward. For example, let's take a movie recommendation system. The idea is first to represent each movie in our catalog as a vector, a numerical representation of a piece of text. Next we also convert the search phrase into a vector. Having done that we enter a whole new realm — a multidimensional space where these vectors replace the original text values. Now through some mathematical techniques, we can determine which movie representations are closest to each other and to our search phrase!

But how do we create such vector representations? We need an AI model trained on vast amounts of data to recognize patterns and effectively convert text phrases into vectors. We also need proper tools to run the model and do the inferences.

All of this and more you’ll learn in this session. We'll try out different data solutions - ClickHouse, OpenSearch, PGVector and others. We'll also explore different models that are available depending on your language preference and programming skills. Or, if you don’t want to run the model locally, what APIs you can use to do the inference for free.

Plenty of demos and a bit of coding for each of the options. This session will be useful for anyone who is intrigued by contextual search and usage of AI, but might find themselves overwhelmed by the complexities to get started.