Capítulo 1: LangChain para RAG

Visão geral do LangChain

LangChain fornece blocos de construção para pipelines de RAG:

• Carregadores e divisores de documentos
• Embeddings e bancos vetoriais
• Retrievers e chains
• LCEL para composabilidade

Chain RAG básica

from langchain.chat_models import ChatOpenAI
from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import Chroma
from langchain.chains import RetrievalQA

# Setup
embeddings = OpenAIEmbeddings()
vectorstore = Chroma.from_documents(documents, embeddings)
llm = ChatOpenAI(model="gpt-4")

# Create chain
qa_chain = RetrievalQA.from_chain_type(
    llm=llm,
    chain_type="stuff",
    retriever=vectorstore.as_retriever()
)

# Query
response = qa_chain.invoke("What is RAG?")
print(response["result"])

LCEL (LangChain Expression Language)

Abordagem moderna e composável:

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

# Define prompt
prompt = ChatPromptTemplate.from_template("""
Answer the question based on the context:

Context: {context}
Question: {question}

Answer:""")

# Build chain with LCEL
chain = (
    {"context": retriever, "question": RunnablePassthrough()}
    | prompt
    | llm
    | StrOutputParser()
)

# Use
response = chain.invoke("What is RAG?")

Tipos de Chain

Tipo	Descrição	Caso de uso
stuff	Coloca todos os documentos no prompt	Contextos pequenos
map_reduce	Processa documentos separadamente e combina	Grandes volumes de dados
refine	Refina a resposta iterativamente	Consultas complexas
map_rerank	Pontua e seleciona o melhor	Múltiplas respostas

Adicionando memória

from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationalRetrievalChain

memory = ConversationBufferMemory(
    memory_key="chat_history",
    return_messages=True
)

chain = ConversationalRetrievalChain.from_llm(
    llm=llm,
    retriever=retriever,
    memory=memory
)

# Mantém o contexto da conversa
chain.invoke({"question": "What is RAG?"})
chain.invoke({"question": "What are its benefits?"})  # Lembra do contexto

---

Próximos passos

Continue para Capítulo 2: LlamaIndex.