Azure OpenAI Service Tutorial

What is Azure OpenAI Service?

Azure OpenAI Service gives you OpenAI models — such as GPT models for chat and text, and embedding models for search — inside Microsoft Azure.

Step 1: Create an Azure OpenAI resource

You need an Azure subscription. Then create the resource from the Azure portal or Azure AI Foundry.

Step 2: Deploy a model

Your code never calls a model directly — it calls a deployment you create from a model.

Step 3: Test in the chat playground

Before writing code, try prompts in the playground to see how the model behaves.

Step 4: Call Azure OpenAI from Python

Install the OpenAI Python SDK, keep your secrets in a .env file and call your deployment.

pip install openai python-dotenv

# .env  (never commit this file)
AZURE_OPENAI_ENDPOINT=https://<your-resource>.openai.azure.com/
AZURE_OPENAI_API_KEY=<your-key>
AZURE_OPENAI_DEPLOYMENT=<your-deployment-name>

import os
from dotenv import load_dotenv
from openai import AzureOpenAI

load_dotenv()

client = AzureOpenAI(
    azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],
    api_key=os.environ["AZURE_OPENAI_API_KEY"],
    api_version="2024-10-21",  # check Microsoft Learn for the latest GA version
)

response = client.chat.completions.create(
    model=os.environ["AZURE_OPENAI_DEPLOYMENT"],  # your deployment name
    messages=[
        {"role": "system", "content": "You are a helpful Azure tutor. Answer in 3 short bullet points."},
        {"role": "user", "content": "What is Azure AI Search used for?"},
    ],
    temperature=0.3,
    max_tokens=300,
)

print(response.choices[0].message.content)

Step 5: Write better prompts

• Set the role in the system message — who the assistant is and what it must not do.
• Be specific about format: bullet points, JSON, word limits.
• Give examples (few-shot) when you need a consistent style.
• Ask for grounded answers — “answer only from the context provided”.
• Lower temperature for factual tasks; raise it for creative ones.

Step 6: Answer from your own documents with RAG

Retrieval-augmented generation lets the model answer using your files instead of guessing.

Security and cost tips