Chapter 11 — Embeddings & Semantic Search

An embedding is the translation of text into a point in space. Texts with nearby meanings become nearby points — and “nearby” becomes a geometry calculation, not an exact-word search.

TL;DR: Embeddings convert text into vectors; searching by similarity between vectors is searching by meaning. It’s the technical foundation of RAG, memory, and nearly all retrieval that feeds an agent.

What is an embedding

An embedding is a vector of numbers (typically 256 to 3072 dimensions) that represents the meaning of text.

async function semanticSearch(tenantId: string, query: string, k = 5) {
  const [vector] = await embed([query]);

  return sql`
    SELECT source, content, 1 - (embedding <=> ${vector}::vector) AS score
    FROM doc_chunks
    WHERE tenant_id = ${tenantId}
    ORDER BY embedding <=> ${vector}::vector
    LIMIT ${k}
  `;
}

When embeddings work well

Embeddings shine in fuzzy meaning:

Works well	Fails
Synonyms and paraphrases	Exact identifiers
Natural language questions	Negation
Related concepts	Precise numbers and dates

The lesson: semantic search and keyword search are complementary.