Deploy Azure OpenAI Service with enterprise patterns for security, scalability, and integration.

Authentication and Security

Managed Identity (Recommended)

from azure.identity import DefaultAzureCredential, get_bearer_token_provider
from openai import AzureOpenAI

# No API keys — uses Azure AD identity
token_provider = get_bearer_token_provider(
    DefaultAzureCredential(),
    "https://cognitiveservices.azure.com/.default"
)

client = AzureOpenAI(
    azure_endpoint="https://your-resource.openai.azure.com",
    azure_ad_token_provider=token_provider,
    api_version="2024-09-01-preview"
)

Private Endpoint Setup

# Create private endpoint for VNet isolation
az network private-endpoint create \
    --name myOpenAIPrivateEndpoint \
    --resource-group myRG \
    --vnet-name myVNet \
    --subnet mySubnet \
    --private-connection-resource-id /subscriptions/.../openai/myOpenAI \
    --group-id account \
    --connection-name myOpenAIConnection

Semantic Kernel Integration

Building AI Applications

from semantic_kernel import Kernel
from semantic_kernel.connectors.ai.open_ai import AzureChatCompletion

kernel = Kernel()

# Add Azure OpenAI service
kernel.add_service(AzureChatCompletion(
    service_id="azure-gpt4o",
    deployment_name="gpt-4o",
    endpoint="https://your-resource.openai.azure.com",
    api_key="your-key"
))

# Create prompts as functions
summarize = kernel.add_function(
    function_name="summarize",
    plugin_name="DocumentProcessing",
    prompt="Summarize this document in 3 bullet points: {{$input}}"
)

# Execute
result = await kernel.invoke(summarize, input=document_text)
print(result)

Multi-Region Load Balancing

import random
from openai import AzureOpenAI

# Multiple regional endpoints for resilience
endpoints = [
    {"url": "https://eastus.openai.azure.com", "key": "key1"},
    {"url": "https://westeurope.openai.azure.com", "key": "key2"},
    {"url": "https://australiaeast.openai.azure.com", "key": "key3"}
]

def get_client():
    # Random selection for load distribution
    endpoint = random.choice(endpoints)
    return AzureOpenAI(
        azure_endpoint=endpoint["url"],
        api_key=endpoint["key"],
        api_version="2024-09-01-preview"
    )

# With retry and fallover
def invoke_with_fallback(prompt, retries=3):
    for attempt in range(retries):
        try:
            client = get_client()
            return client.chat.completions.create(
                model="gpt-4o",
                messages=[{"role": "user", "content": prompt}]
            )
        except Exception as e:
            if attempt == retries - 1:
                raise