What is Semantic Search?

What is Semantic Search?

Semantic search finds results based on meaning and intent, not just exact keyword matches. It uses vector embeddings to represent text as high-dimensional vectors, then measures similarity between vectors to surface conceptually related content — even when different words are used.

How It Works

1. Embed the query and all documents using an embedding model (e.g.,
   textCopy

   text-embedding-3-small

   )
2. Index document vectors in a vector store (FAISS, Pinecone, Chroma)
3. Search by computing cosine similarity or dot product between query and stored vectors
4. Return the top-K most similar documents

Semantic Search vs Keyword Search

Code Example

pythonCopy
from sentence_transformers import SentenceTransformer
import faiss
import numpy as np

model = SentenceTransformer("all-MiniLM-L6-v2")

# Document corpus
docs = [
    "How to train a neural network",
    "Best practices for LLM fine-tuning",
    "Introduction to gradient descent",
]

# Embed and index documents
doc_embeddings = model.encode(docs)
index = faiss.IndexFlatL2(doc_embeddings.shape[1])
index.add(np.array(doc_embeddings))

# Query by meaning, not keywords
query = "How do I teach a model new skills?"
query_vec = model.encode([query])

distances, indices = index.search(query_vec, k=2)
for i in indices[0]:
    print(docs[i])
# Output: "Best practices for LLM fine-tuning", "How to train a neural network"

Use Cases

• RAG pipelines — retrieve relevant documents for LLM context grounding
• Enterprise search — search internal knowledge bases by meaning
• Recommendation systems — find similar products, articles, or users
• Duplicate detection — identify near-duplicate content semantically

Hybrid Search (Best of Both Worlds)

Combine semantic + keyword search using Reciprocal Rank Fusion (RRF):

pythonCopy
# LangChain hybrid retriever example
from langchain.retrievers import EnsembleRetriever

retriever = EnsembleRetriever(
    retrievers=[bm25_retriever, faiss_retriever],
    weights=[0.4, 0.6]  # 40% keyword, 60% semantic
)

Best Practice: Use hybrid search in production RAG systems — semantic handles meaning, keyword handles exact matches (product IDs, names, codes).

Popular tools: FAISS, Pinecone, Weaviate, Chroma, Qdrant