Retrieval-Augmented Generation with LangChain

This course's earlier ancestor, plain RAG, is common enough that LangChain provides ready-built pieces for every step: loading documents, splitting them, embedding, storing, and retrieving. This lesson wires those pieces into a chain instead of hand-writing each step.

Loading and splitting documents

from langchain_text_splitters import RecursiveCharacterTextSplitter

text = 'LangChain is a framework for building applications with LLMs...'
splitter = RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=20)
chunks = splitter.split_text(text)
print(len(chunks), chunks[0])

RecursiveCharacterTextSplitter tries to split on paragraph breaks first, falling back to sentences, then words, so chunks stay as semantically coherent as possible instead of cutting mid-sentence at a fixed character count. chunk_overlap repeats a few characters between adjacent chunks so a fact split across a chunk boundary still appears whole in at least one chunk.

Embedding and storing in a vector store

from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS

embeddings = OpenAIEmbeddings(model='text-embedding-3-small')
vectorstore = FAISS.from_texts(chunks, embeddings)
retriever = vectorstore.as_retriever(search_kwargs={'k': 3})

This needs a real API key to compute embeddings, so it's read-only here. FAISS.from_texts embeds every chunk and stores the resulting vectors for similarity search, .as_retriever() wraps the vector store as a Runnable, so it fits directly into an LCEL chain, search_kwargs={'k': 3} returns the 3 most similar chunks for a given query.

Wiring retrieval into a chain

from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough

prompt = ChatPromptTemplate.from_template(
    'Answer the question using only this context:\n{context}\n\nQuestion: {question}'
)

def format_docs(docs):
    return '\n\n'.join(doc.page_content for doc in docs)

rag_chain = (
    {'context': retriever | format_docs, 'question': RunnablePassthrough()}
    | prompt
    | model
    | StrOutputParser()
)

answer = rag_chain.invoke('What is LangChain?')

Read-only, needs a real model and vector store. The dictionary {'context': ..., 'question': ...} runs both branches in parallel on the same input: retriever | format_docs turns the raw question into retrieved, formatted context, while RunnablePassthrough() just forwards the original question unchanged, both results land in prompt's two placeholders.