Building RAG Systems ThatActually Work in Production

# Simplified RAG pipeline
from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import Pinecone
from langchain.chat_models import ChatOpenAI
from langchain.chains import RetrievalQA

# 1. Initialize components
embeddings = OpenAIEmbeddings(model="text-embedding-3-large")
vectorstore = Pinecone.from_existing_index("my-index", embeddings)
llm = ChatOpenAI(model="gpt-4-turbo", temperature=0)

# 2. Create retrieval chain
qa_chain = RetrievalQA.from_chain_type(
    llm=llm,
    chain_type="stuff",
    retriever=vectorstore.as_retriever(
        search_type="mmr",  # Maximal Marginal Relevance
        search_kwargs={"k": 5, "fetch_k": 20}
    ),
    return_source_documents=True,
)

# 3. Query
result = qa_chain({"query": "What is our refund policy?"})
print(result["result"])
print(result["source_documents"])

from langchain.text_splitter import RecursiveCharacterTextSplitter

# Production-tested settings
splitter = RecursiveCharacterTextSplitter(
    chunk_size=1000,          # Sweet spot for most use cases
    chunk_overlap=200,        # 20% overlap prevents context loss
    length_function=len,
    separators=[
        "\n\n",             # Paragraphs first
        "\n",                # Then lines
        ". ",                 # Then sentences
        " ",                  # Then words
        ""                    # Finally characters
    ],
    is_separator_regex=False,
)

# For code: use language-aware splitter
from langchain.text_splitter import Language, RecursiveCharacterTextSplitter

code_splitter = RecursiveCharacterTextSplitter.from_language(
    language=Language.PYTHON,
    chunk_size=1500,
    chunk_overlap=200,
)

# Pinecone hybrid search
from pinecone import Pinecone

pc = Pinecone(api_key="...")
index = pc.Index("my-index")

# Query with both vector and sparse (BM25) embeddings
results = index.query(
    vector=dense_embedding,      # From OpenAI/Cohere
    sparse_vector=sparse_embedding,  # BM25/SPLADE
    top_k=10,
    include_metadata=True,
    filter={"category": "support"},  # Metadata filtering
)

# Weighted combination (alpha = vector weight)
# alpha=1.0 = pure vector, alpha=0.0 = pure keyword
# alpha=0.7 is a good starting point

import cohere

co = cohere.Client("...")

# First: retrieve top 50 with vector search
candidates = vectorstore.similarity_search(query, k=50)

# Then: rerank to top 5
reranked = co.rerank(
    model="rerank-english-v3.0",
    query=query,
    documents=[doc.page_content for doc in candidates],
    top_n=5,
)

# Use reranked results
final_docs = [candidates[r.index] for r in reranked.results]