langgraph-dev-basic-rag

Basic RAG in LangGraph

RAG (Retrieval-Augmented Generation) grounds LLM responses in retrieved documents, improving factual accuracy and enabling knowledge-base QA.

RAG Pipeline

Index Phase: Load -> Split -> Embed -> Store Query Phase: Embed -> Search -> Retrieve -> Augment -> Generate

Installation

Using UV (Recommended)

uv add langchain-community langchain-core langchain-anthropic
uv add langchain-openai  # For embeddings (Anthropic doesn't provide embeddings)
uv add langchain-text-splitters
uv add faiss-cpu  # or faiss-gpu for CUDA support
uv add langgraph

Using pip

pip install langchain-community langchain-core langchain-anthropic
pip install langchain-openai langchain-text-splitters
pip install faiss-cpu langgraph

Note: Anthropic does not provide embedding models. Use OpenAI, Voyage AI, or HuggingFace for embeddings.

Implementation

from langchain_community.document_loaders import TextLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_core.documents import Document
from typing import List

# 1. Load documents
loader: TextLoader = TextLoader("documents.txt")
documents: List[Document] = loader.load()

# 2. Split into chunks
text_splitter: RecursiveCharacterTextSplitter = RecursiveCharacterTextSplitter(
    chunk_size=1000,
    chunk_overlap=200
)
splits: List[Document] = text_splitter.split_documents(documents)

# 3. Create embeddings
embeddings: OpenAIEmbeddings = OpenAIEmbeddings()

# 4. Create vector store
vectorstore: FAISS = FAISS.from_documents(splits, embeddings)

# 5. Create retriever
retriever = vectorstore.as_retriever(
    search_type="similarity",
    search_kwargs={"k": 3}
)

# Optional: Save/Load vector store for reuse
vectorstore.save_local("faiss_index")
# To load: vectorstore = FAISS.load_local("faiss_index", embeddings, allow_dangerous_deserialization=True)