Embeddings
Numeric representations of text (or images) where similar meaning lands close together — the mathematics that lets a system search by meaning instead of keywords.
Production AI is a system: context, tools, permissions, traces, evals, and feedback loops around the model.
What it is
An embedding model turns a piece of text into a vector — a long list of numbers — positioned so that texts with similar meaning end up near each other. 'Refund my order' and 'I want my money back' share almost no words but land close in embedding space. That property is what powers semantic search.
Why it matters
Embeddings are the bridge between how people phrase things and how your data is written. They power retrieval for assistants, duplicate detection, clustering of support themes, and recommendation. When AI answers feel like they 'understood the question,' embeddings did the understanding.
How it works
Documents are chunked, each chunk is embedded once and stored in a vector index (pgvector in our stack); at question time the query is embedded and the nearest chunks are retrieved. Embeddings go stale when content changes, so refresh runs on a documented cadence, and the embedding model version is tracked — switching models means re-embedding the corpus.
Related resources
How an AI agent finds the right document, chunk, or row to ground its answer in — and why the part that matters is the pipeline around the database, not the database itself.
Search that matches meaning instead of exact words — ask in your own phrasing and find the document that answers, even when it uses different vocabulary.
Splitting documents into retrieval-sized pieces — small enough to search precisely, large enough to keep their meaning — before embedding them for search.
The Postgres extension that adds vector types and similarity indexes to the database you already run — the pragmatic default for embedding search below very large scale.