Semantic Search
AISemantic search is finding content by meaning rather than by exact keyword match. A query like "when is payment due" can match passages about "due date," "pay by," or "invoice terms" because embeddings capture semantic similarity. DocLD implements semantic search via vector search over embedded chunks in Pinecone.
vs Full-Text Search
| Approach | Matches by | Best for |
|---|---|---|
| Semantic | Meaning (embeddings) | Questions, paraphrases, concepts |
| Full-text | Keywords (lexical) | Exact terms, IDs, names |
RAG in DocLD uses semantic search (with optional reranking) to retrieve chunks before LLM completion. Hybrid search combines semantic and lexical search.
Related Concepts
Semantic search is implemented with vector search and embeddings. It powers retrieval for RAG and knowledge base chat.