What are best practices for building production-grade Gemini API applications?

# Example production-hardened Gemini call:
from google import genai
from google.genai import types
from google.api_core import exceptions
import time, logging

logger = logging.getLogger("gemini-app")
client = genai.Client()

GEMINI_MODEL = "gemini-2.5-flash-001"  # PINNED - never use -latest

def production_generate(prompt: str, user_id: str) -> str:
    for attempt in range(4):
        try:
            response = client.models.generate_content(
                model=GEMINI_MODEL,
                contents=prompt,
                config=types.GenerateContentConfig(
                    max_output_tokens=2048,   # cap output
                    temperature=0.2,          # low for predictability
                    safety_settings=[
                        types.SafetySetting(
                            category="HARM_CATEGORY_DANGEROUS_CONTENT",
                            threshold="BLOCK_MEDIUM_AND_ABOVE"
                        )
                    ]
                )
            )
            if not response.candidates:
                raise ValueError(f"Blocked: {response.prompt_feedback.block_reason}")
            logger.info({"user": user_id, "tokens": response.usage_metadata.total_token_count})
            return response.text
        except exceptions.ResourceExhausted:
            time.sleep(2 ** attempt)
    raise RuntimeError("Max retries exceeded")