What an embedding is
An embedding is a list of numbers — a vector — that a model produces to represent a piece of content. Content with similar meaning ends up with vectors that are close together, so “refund request” and “I want my money back” land near each other even though they share no words.
What they are used for
Embeddings make meaning searchable and comparable:
- Semantic search over documents, tickets or knowledge bases.
- Retrieval for AI assistants: finding the right passages to give a model as context.
- Grouping similar items, and spotting duplicates or outliers.
- Matching images to images, or images to text descriptions.
Practical cautions
Vectors inherit the permissions problem of the content they came from: if a user should not see a document, they should not retrieve its passages either. Re-embed when the source changes, and keep the link back to the original record so answers can be checked.