LangChain with Vector Databases

What vector databases are, how they work, a comparison of the popular ones, LangChain's common VectorStore interface and examples with FAISS, Chroma and Qdrant.

12Retrieval Augmented Generation (RAG) in LangChain 15Embeddings and Semantic Search 16Retrieval-Augmented Generation (RAG) in LangChain

13.1

What are Vector Databases?

Vector databases store text, images or other data as numerical vectors (embeddings).
They enable fast similarity search to find semantically relevant data.
Essential for RAG applications.
Popular vector databases: FAISS, Chroma, Qdrant, Weaviate, Pinecone, Milvus, etc.
LangChain provides easy integration with these vector databases.

13.2

Why Use Vector Databases?

Find relevant information based on meaning (semantic search).
Works better than keyword search.
Handles large amounts of unstructured data.
Improves accuracy of LLM responses.
Supports filtering (metadata-based search).
Scalable and high-performance for enterprise use.
Useful for document search, knowledge bases, chatbots and RAG systems.

13.3

How a Vector Database Works

DocumentsPDF, TXT, etc.

Split TextChunks

Create EmbeddingsUsing embedding model

Store in Vector Database

Similarity SearchIn vector DB

Relevant ChunksTo LLM

Final ResponseTo user

At query time the user query is converted to an embedding and the top-K relevant documents are returned.

13.4

Popular Vector Databases Comparison

Vector DB	Type	Key Features	Use Case
FAISS	Local (library)	Fast, simple, in-memory	Local development, small to medium data
Chroma	Local / server	Easy to use, persistent storage	Prototyping, open-source projects
Qdrant	Server / cloud	High performance, filtering, payload	Production, scalable applications
Pinecone	Cloud	Managed service, scalable	Enterprise, production use
Weaviate	Server / cloud	Hybrid search, metadata filtering	Production, semantic search
Milvus	Server / cloud	High scalability, distributed	Large-scale enterprise applications

13.5

LangChain VectorStore Abstractions

LangChain provides a common VectorStore interface for different databases.
You can easily switch between databases by changing a few lines of code.
Supports adding documents, similarity search and metadata filtering.
Integrates with embedding models and retrievers.
Works seamlessly with RAG chains.

add_documents(): Add documents to the vector store
similarity_search(): Search similar documents
similarity_search_with_score(): Get results with scores
as_retriever(): Create a retriever for RAG
delete(): Delete documents
get(): Get documents by id

13.6

Example 1: FAISS

python

from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import CharacterTextSplitter
from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import FAISS

# 1. Load and split document
loader = PyPDFLoader("data/company_policy.pdf")
docs = loader.load()
splitter = CharacterTextSplitter(chunk_size=500, chunk_overlap=50)
chunks = splitter.split_documents(docs)

# 2. Create embeddings
embeddings = OpenAIEmbeddings()

# 3. Create FAISS vector store
vectorstore = FAISS.from_documents(chunks, embeddings)

# 4. Similarity search
results = vectorstore.similarity_search("What is the leave policy?", k=3)
for doc in results:
    print(doc.page_content)
    print("Source:", doc.metadata)

13.7

Example 2: Chroma

python

from langchain.document_loaders import DirectoryLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import Chroma

# Load documents
loader = DirectoryLoader("data/")  # folder with multiple files
docs = loader.load()
splitter = RecursiveCharacterTextSplitter(chunk_size=500,
                                          chunk_overlap=50)
chunks = splitter.split_documents(docs)

# Create embeddings
embeddings = OpenAIEmbeddings()

# Create Chroma vector store (persistent storage)
vectorstore = Chroma.from_documents(
    documents=chunks,
    embedding=embeddings,
    persist_directory="chroma_db"
)

# Search
results = vectorstore.similarity_search("What is KYC process?", k=3)
for doc in results:
    print(doc.page_content)
    print("Source:", doc.metadata)

13.8

Example 3: Qdrant

python

from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import Qdrant

embeddings = OpenAIEmbeddings()
vectorstore = Qdrant.from_documents(
    documents=chunks,
    embedding=embeddings,
    url="http://localhost:6333",   # Qdrant local server
    collection_name="company_docs"
)

13.9

Key Takeaways

Vector databases enable semantic search for better results.
LangChain provides a unified interface for multiple vector stores.
Choose the right database based on scale, cost and use case.
Use proper chunking and embeddings for accurate search.
Works seamlessly with RAG and agents.
Supports metadata filtering and hybrid search.
Essential for building scalable LLM applications.