The Yandex B2B Tech team has implemented hybrid search in the YDB database. The technology combines exact match search and semantic search, allowing it to simultaneously consider keywords, document numbers, and query context.
The solution is designed for IT products that require working with different types of data. Possible scenarios include recommendation systems for marketplaces and social networks, chatbots, and other AI services. For example, when searching for information on an insurance policy, the system will be able to simultaneously find the exact document number and understand the description of the insured event.
YDB became the first Russian database with built-in hybrid search. The technology is available in the enterprise edition version 26.3 for on-premises deployments and in Managed Service for YDB.
Previously, for such tasks, companies had to use separate systems for the database, full-text, and vector search, and then synchronize information between them. In YDB, data and both types of indexes are located in one system and are updated within a single transaction.
According to Andrey Fomichev, YDB Technical Director, the combination of two approaches allows for considering both exact terms and codes, as well as semantic connections between formulations. The main engineering challenge during development was combining and ranking the results of full-text and vector search.