semantic search

Semantic search finds relevant content based on meaning similarity rather than exact keyword matches. **How it works**: Convert query and documents to dense vector embeddings, compute similarity (cosine, dot product), return nearest neighbors. **Advantages over keyword search**: Understands synonyms ("car" finds "automobile"), handles paraphrase ("how to fix" matches "repair instructions"), captures conceptual relationships. **Embedding models**: OpenAI text-embedding-3, Cohere embed, sentence-transformers, BGE, E5. **Vector databases**: Pinecone, Weaviate, Milvus, Qdrant, Chroma, pgvector. **Implementation**: Chunk documents → embed → store in vector DB → embed query → retrieve top-k similar. **Optimization**: Approximate nearest neighbor (ANN) algorithms (HNSW, IVF) for scale. **Limitations**: May miss exact matches keywords would find, embedding quality varies, struggles with rare terms. **Best practices**: Use hybrid search (semantic + keyword), tune chunk size, consider domain-specific embeddings. Foundation of modern RAG systems.