Skip to content

Use Azure OpenAI in Foundry Models to Generate Content

This section of the Microsoft AI-102: Designing and Implementing a Microsoft Azure AI Solution exam covers using Azure OpenAI models in Foundry to generate text, code, and images. Below are study notes for each sub-topic, with links to Microsoft documentation, exam tips, and key facts

Provision an Azure OpenAI in Foundry Models Resource

📖 Docs: Provision Azure OpenAI resources

Overview

  • To use Azure OpenAI in Foundry, you must provision a resource in the Azure portal
  • Requires approval from Microsoft due to responsible AI usage guidelines
  • Must be linked into a Foundry hub/project

Key Points

  • Choose the right region (not all regions support OpenAI)
  • Requires quota approval
  • Controlled with RBAC and subscription limits

Limits

Azure OpenAI is not available in all regions and requires Microsoft approval

Select and Deploy an Azure OpenAI Model

📖 Docs: Foundry Models sold directly by Azure

Overview

  • Models must be deployed to a resource before use
  • Supported models:
    • GPT-4 Turbo (text + multimodal)
    • GPT-3.5 Turbo
    • Embeddings models
    • DALL·E (image generation)
    • Whisper (speech-to-text)

Key Points

  • Choose the correct model based on use case
  • Deployment includes selecting capacity and model version
  • Endpoint and key are created for API access

Exam Tip

Be ready to match task → model choice in questions

Submit Prompts to Generate Code and Natural Language Responses

📖 Docs: Completions API

Overview

  • Prompts submitted to deployed models using REST API or SDK
  • Supports tasks such as:
    • Summarization
    • Code generation
    • Question answering
    • Chat completion

Key Points

  • Token usage = input + output tokens
  • System messages help define assistant behavior
  • Can implement temperature and top_p to control creativity

Code Snippet

from openai import AzureOpenAI

client = AzureOpenAI(api_key="KEY", api_version="2023-07-01-preview")
resp = client.chat.completions.create(
    model="gpt-35-turbo",
    messages=[{"role":"system","content":"You are a helpful assistant"},
              {"role":"user","content":"Write Python code for Fibonacci"}]
)
print(resp.choices[0].message["content"])

Use the DALL-E Model to Generate Images

📖 Docs: Generate images with Azure OpenAI

Overview

  • DALL·E generates images from natural language prompts
  • Supports:
    • New images
    • Image edits
    • Variations

Key Points

  • Output formats: PNG, JPEG
  • Size options: 256×256, 512×512, 1024×1024
  • Counts tokens separately from text models

Exam Tip

Keywords like generate images or create visuals → DALL·E

Integrate Azure OpenAI into Your Own Application

📖 Docs: Integrate Azure OpenAI

Overview

  • Applications integrate via:
    • REST APIs
    • Azure OpenAI SDKs
    • Foundry SDK for orchestration
  • Authentication with API keys or Azure AD

Key Points

  • Endpoints are region-specific
  • Secure keys in Azure Key Vault
  • Common integrations:
    • Chatbots
    • Copilot-style apps
    • Knowledge-grounded solutions

Use Case

Integrating GPT with a customer support portal

Use Large Multimodal Models in Azure OpenAI

📖 Docs: GPT-4o or GPT-4o Mini with Microsoft Fabric for Image Data

Overview

  • GPT-4 Turbo supports multimodal input (text + images)
  • Enables use cases such as:
    • Image captioning
    • Visual reasoning
    • Analyzing diagrams or screenshots

Key Points

  • Images are sent as part of the prompt
  • Output is still text
  • Token usage can increase with multimodal prompts

Exam Tip

If the question involves reasoning over images, the answer is GPT-4 Turbo with vision

Implement an Azure OpenAI Assistant

📖 Docs: Getting started with Azure OpenAI Assistants

Overview

  • Assistants API enables long-running AI agents
  • Features:
    • Tool integration (code interpreter, retrieval)
    • Persistent threads and memory
    • Multi-turn conversations

Key Points

  • Assistants differ from single-shot prompts
  • Useful for copilots and interactive apps
  • Configured in Foundry or via SDK

Use Case

A coding assistant that maintains state across multiple prompts

Quick‑fire revision sheet

  • 📌 Must provision Azure OpenAI resource with Microsoft approval
  • 📌 Models must be deployed before use
  • 📌 GPT models → text/code, DALL·E → images, Whisper → speech
  • 📌 Token usage = input + output tokens
  • 📌 GPT-4 Turbo supports multimodal inputs
  • 📌 Assistants API = stateful, tool-augmented AI agents