Embeddings
Numerical vector representations of text (or other content) that capture semantic meaning, enabling similarity search and retrieval.
In detail
An embedding is a fixed-size array of floating-point numbers produced by a model when it encodes a piece of text, image or other content. Semantically similar content produces similar (close in vector space) embeddings. This property is what makes similarity search work: given a query, you embed it, then find the closest stored embeddings in a vector database and retrieve their associated content. Embedding models include OpenAI's text-embedding series, Cohere's Embed models, and open-source options like Nomic Embed. Embeddings are the input layer for retrieval-augmented generation and semantic search.
Why it matters for Australian business
For Australian businesses embeddings are most relevant when building internal knowledge bases, document search tools or AI assistants that need to find relevant content across a large corpus. The choice of embedding model and vector database affects both the quality of retrieval and where data lives. Australian data residency concerns apply here too: embedding models hosted offshore process the text you send through them.