Text → vectors for semantic search and retrieval
When searching an external library of 10 million books, the scholar doesn't read every sentence sequentially. An embedding model pre-computes a dense fingerprint vector for every document chunk. When you ask a question, your query is turned into a matching fingerprint vector ($\mathbf{q}$), and finding the top passages is simply pointing the scholar's index finger at the nearest coordinates ($\mathbf{q} \cdot \mathbf{d}_i$)!