What is retrieval-augmented generation (RAG) and how do you implement it with OpenAI?

from openai import OpenAI

client = OpenAI()

# ---- FULL RAG PIPELINE ----

# Step 1: Embed and index your documents (do this once)
def embed_documents(texts: list[str]) -> list[list[float]]:
    response = client.embeddings.create(
        model="text-embedding-3-large",
        input=texts,
    )
    return [item.embedding for item in response.data]

docs = [
    "Our refund policy: returns accepted within 30 days with receipt.",
    "API rate limits: 3500 RPM for Tier 2 customers.",
    "Python 3.12 introduced GIL opt-out via Py_GIL_DISABLED=1.",
]
doc_embeddings = embed_documents(docs)
# Store doc_embeddings + docs in a vector database (Pinecone, Weaviate, pgvector...)

# Step 2: At query time, retrieve relevant chunks
user_query = "Can I return a product after 2 months?"
query_embedding = embed_documents([user_query])[0]
# similarity_search(query_embedding) -> returns top-k relevant docs
relevant_docs = ["Our refund policy: returns accepted within 30 days with receipt."]

# Step 3: Generate with context
response = client.responses.create(
    model="gpt-5.5",
    instructions="Answer based only on the provided context. If unsure, say so.",
    input=f"Context:\n{chr(10).join(relevant_docs)}\n\nQuestion: {user_query}",
)
print(response.output_text)
# "Based on our policy, returns are only accepted within 30 days with a receipt,
# so a return after 2 months would not be eligible."