Embeddings and Semantic Search

What embeddings are, popular embedding models, the workflow, how semantic search and similarity metrics work, advanced techniques and use cases.

13LangChain with Vector Databases 14Document Loaders and Text Splitters 16Retrieval-Augmented Generation (RAG) in LangChain

15.1

What are Embeddings?

Embeddings are numerical vector representations of text, capturing semantic meaning.
They convert text (documents, queries) into dense vectors (e.g., 1536 dimensions) using embedding models.
Similar content will have similar vectors, enabling semantic search, RAG and AI applications.

15.2

Popular Embedding Models

OpenAI Embeddings - text-embedding-3-small
Cohere Embeddings - multilingual models
Hugging Face - sentence-transformers
Local models - BGE, all-MiniLM, Instructor models
Multimodal models - for text, images, code, etc.

15.3

Embedding Workflow in LangChain

Raw DocumentsPDF, web, text

Text SplitterSplit into chunks

Embedding ModelCreate embeddings

Vector DatabaseStore vectors

Semantic SearchRetrieve relevant chunks

15.4

How Semantic Search Works

1The user query is converted into an embedding vector.
2The vector is compared with stored document vectors (using similarity search).
3The top-k most relevant chunks are retrieved and used by the LLM to generate accurate answers.

15.5

Similarity Metrics

Metric	Description	Use Case
Cosine Similarity	Measures angle between vectors	Most common
Euclidean Distance	Measures actual distance	When magnitude matters
Dot Product	Simple and efficient	Suitable for large-scale search

15.6

Implementing Embeddings in LangChain

python

from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import FAISS

# Create embeddings
embeddings = OpenAIEmbeddings()
vectorstore = FAISS.from_documents(docs, embeddings)

15.7

Semantic Search with LangChain

python

from langchain.vectorstores import FAISS
from langchain.chains import RetrievalQA

retriever = vectorstore.as_retriever(search_kwargs={
    "k": 3
})
qa_chain = RetrievalQA.from_chain_type(llm=llm,
                                       retriever=retriever)

result = qa_chain.run("What is LangChain?")
print(result)

15.8

Advanced Techniques

Hybrid Search - combine keyword + vector search.
Metadata Filtering - filter results by source, date, tags.
Re-ranking - improve results using cross-encoders.
Multi-vector Retrieval - use multiple embeddings (e.g., title + content).
Semantic Chunking - split content based on meaning, not just size.

15.9

Use Cases of Embeddings

RAG applications - retrieve relevant context for the LLM.
Semantic search - intelligent search in docs and products.
Recommendation systems - find similar items.
Chatbots and assistants - answer using your data.
Multimodal search - search across text, images and code.

15.10

Example: RAG with Embeddings

PDFCompany docs

Split & Embed

Vector DBFAISS / Qdrant

Retrieve Relevant Docs

LLMGive accurate answer

15.11

Key Takeaways

Embeddings bridge the gap between text and machine understanding.
Semantic search helps find the most relevant information efficiently.
LangChain makes it easy to generate embeddings, store them in vector databases and use them for powerful RAG applications.
Choose the right model, metric and database based on your use case and scale.