Translated with AI from the Russian original.
The technology is available in the new version of the YDB database management system both in the cloud and locally - using the on-premises model.
Vector search is a technology that can search not by keywords, but by meaning. According to the developers, it will help companies that develop AI-based products to, for example, improve the accuracy of product recommendations or the quality of AI assistant responses. Vector search technology is already used in the "Alice" AI assistant with the "Pro" option.
Yandex explained:
Vector technology involves representing the data being searched as numerical vectors, or embeddings. By comparing vectors, it is possible to determine the semantic proximity of the objects corresponding to them. This approach provides a number of advantages: from the ability to establish non-obvious connections between objects and search through heterogeneous data - for example, texts, pictures, videos and audio files - to resistance to errors and typos.
The YDB DBMS offers two versions of vector search: exact and approximate. Exact search provides the best result, but requires complex calculations if there is a lot of data. Approximate search allows you to search through billions of vectors in tens to hundreds of milliseconds - even if the vector corpus does not fit into RAM.
YDB is developing vector search with the support of the Yandex Technological Initiatives Fund - YATF.
