Embeddings in practice

Every search in this course starts with one decision: which function turns a clause, a register entry or a question into a vector. That function decides what "similar" means for every query your agents will ever run, and changing it later means re-embedding everything. Treat the embedding model like a database engine you'll live with for years: pick it on evidence from your own documents, then size the index so it's cheap enough to keep.

The theory of embeddings (vectors, dot products, cosine similarity) is covered in Tokens and embeddings. This lesson is about practice: the model, the input conventions, and the three knobs that set memory and cost, which are dimensions, truncation and precision.