AI models operate on numerical data (vectors). To make content usable by these models, it must be first converted it into vector form — known as embeddings.
This requires breaking each document into meaningful semantic units (e.g., paragraphs, illustrations, images) and using an embedding model to encode these units into vectors.