How do you build a Retrieval-Augmented Generation (RAG) pipeline with the Gemini API?

from google import genai
from google.genai import types
import numpy as np

client = genai.Client()

# APPROACH 1: Built-in file_search (simplest - Google manages everything)
# 1. Upload documents:
uploaded = client.files.upload(file="knowledge-base.pdf")

# 2. Query against them (no vector DB needed):
response = client.models.generate_content(
    model="gemini-3.5-flash",
    contents="What is our refund policy?",
    config=types.GenerateContentConfig(
        tools=[types.Tool(retrieval=types.Retrieval(
            vertex_ai_search=types.VertexAISearch(datastore="projects/my-project/...")
        ))]
    )
)
print(response.text)

# APPROACH 2: Custom RAG with Gemini embeddings
def embed(texts: list[str]) -> list[list[float]]:
    result = client.models.embed_content(
        model="gemini-embedding-exp-03-07",
        contents=texts,
        config=types.EmbedContentConfig(task_type="RETRIEVAL_DOCUMENT"),
    )
    return [e.values for e in result.embeddings]

# Index your documents:
docs = ["Our policy is...", "For returns, contact..."]
doc_embeddings = embed(docs)  # store in Pinecone/pgvector/Weaviate

# At query time:
query_embedding = client.models.embed_content(
    model="gemini-embedding-exp-03-07",
    contents="Can I return a product?",
    config=types.EmbedContentConfig(task_type="RETRIEVAL_QUERY"),
).embeddings[0].values

# Find most similar docs (cosine similarity):
similarities = [np.dot(query_embedding, d) /
                (np.linalg.norm(query_embedding) * np.linalg.norm(d))
                for d in doc_embeddings]
top_idx = int(np.argmax(similarities))

# Generate with context:
response = client.models.generate_content(
    model="gemini-3.5-flash",
    contents=f"Context: {docs[top_idx]}\n\nQuestion: Can I return a product?",
)