Glossary · AI & ML

Embeddings for Databases

Embeddings for databases are numerical vector representations of schema elements–table names, column names, descriptions–that let AI systems match user questions to the right data by meaning instead of exact wording.

An embedding turns text into a list of numbers positioned so that similar meanings land close together: "revenue", "sales", and "income" cluster near each other even though they share no letters. Applied to a database, embeddings give every table and column a semantic address. When a question arrives, the system embeds it and looks up which schema elements live nearby–instantly narrowing 500 tables to the 5 that matter. The business payoff is that people can use their own vocabulary instead of memorizing the database's, which is most of what "ask your data anything" actually requires.

Fig — every answer ships with the tables, rows and SQL behind it.