Semantic search
Searching by meaning rather than exact keywords, by comparing the vector embedding of a query against stored embeddings and returning the closest matches.
Semantic search finds content by meaning instead of exact word matches. The query and the stored text are both turned into embeddings — numeric vectors that capture meaning — and the system returns the items whose vectors sit closest to the query. A search for “customer refund policy” can surface a passage that says “how we handle returns,” even with no shared keywords.
For transcription, this is what makes recordings queryable by concept. Each transcript chunk is embedded and stored in a vector database, so you can ask a question and retrieve the relevant moments across hours of audio rather than scrolling for exact phrases. It is also the retrieval step that grounds RAG systems and AI agents in your own recorded content.