Embeddings – text as numbers

What is an embedding?

An embedding is a list of numbers (a vector) that represents the meaning of a piece of text – a word, a sentence, a whole paragraph – in a multi-dimensional space. A dedicated embedding model produces these numbers, not the language model that later answers your questions.

Why this is useful

Texts with similar meaning get vectors that sit close together in number space. "Dog" and "puppy" sit close together, "dog" and "tax return" sit far apart. A computer can calculate this closeness (e.g. with cosine similarity) without "understanding" language itself – plain math is enough.

What embeddings are used for

• Semantic search: find documents that match in meaning, even when no word matches exactly
• RAG: the foundation for finding relevant text snippets for a request
• Clustering: automatically group similar texts
• Recommendations: "similar articles" features

Where embeddings get stored

At larger scale, embeddings end up in a vector database (e.g. Pinecone, Weaviate, or the Postgres extension pgvector), which is optimized for fast similarity search across millions of vectors.