Embedding
An embedding is a vector, a list of numbers, that represents the meaning of a piece of text, an image, or another kind of data.
Category
Industry concept
See all terms
What is an embedding?
An embedding is a vector, a list of numbers, that represents the meaning of a piece of text, an image, or another kind of data. A model produces this vector by mapping the input into a space where items with similar meaning end up close together and items with different meaning end up far apart.
This solves a basic problem: computers compare numbers easily but have no native way to compare meaning. Turning "dog" and "puppy" into vectors that sit near each other lets a machine treat them as related without having to write a rule to define this.
How does an embedding differ from a vector database?
An embedding is a single vector produced for one piece of data by a model. A vector database is a separate system that stores millions of those vectors and searches them by similarity, the index, not the representation itself. These are mutually dependent: an embedding with nowhere to search is only a number, and a vector database with nothing embedded has nothing to search.
How are embeddings created?
A model trained for this purpose reads the input and outputs a fixed-length vector, often hundreds or thousands of numbers, positioned so that inputs with related meaning land close together by a distance measure. It learns this positioning during training, seeing large amounts of data and adjusting its output so related items move closer together and unrelated ones move apart.
Embeddings only stay useful if what they represent stays accurate. Qontext keeps the source knowledge behind any embedding structured and current in the context repository, so a search that surfaces a document reflects what that document actually says today, not an outdated snapshot from whenever it was last embedded.