Dense vector representations of text where semantically similar content maps to nearby points in high-dimensional space. "How do I deploy?" and "What's the deployment process?" are far apart in string distance but close in embedding space. This is the magic that makes RAG work — you search by meaning, not by keywords.