What is a vector embedding?

A vector embedding turns text — a word, a sentence, a document, a column name — into a point in a high-dimensional space, expressed as a list of numbers. The trick is that the placement captures meaning: "revenue" and "total sales" land near each other even though they share no letters, while "revenue" and "weather" land far apart. That lets software search by semantic similarity — finding what you meant, not just what you literally typed — which powers semantic search and much of modern AI retrieval.

iDBQuery uses embeddings behind the scenes to bridge the gap between how people ask questions and how data is actually named and stored:

• Matching questions to your schema. When you ask about "customers who churned last quarter," embeddings help iDBQuery find the relevant tables and columns even if they're labelled with cryptic codes.
• Powering the semantic layer. The meanings captured for obscure fields help the model reason in your business's terms.
• Searching documents. For PDFs and folders of files, embeddings help retrieve the passages relevant to your question.

Crucially, embeddings are how iDBQuery *understands and locates* the right data — but the actual answer to a structured question is then computed by running real SQL against those sources and cited back to the exact rows. So embeddings provide the semantic matchmaking, and iDBQuery pairs it with verifiable querying rather than leaving similarity to do all the work.