1: Plan and manage an Azure AI solution

Learning Objectives

By the end of this chapter, you will be able to:

• Select the appropriate service for a generative AI solution, including choosing AI models and protecting account keys.
• Select the appropriate service for a computer vision solution, including deployment options and authentication for a Microsoft Foundry Service resource.
• Select the appropriate service for a natural language processing solution, including SDK and API selection plus content moderation.
• Select the appropriate service for a speech solution, including identifying default endpoints and configuring responsible AI insights and content safety.
• Select the appropriate service for an information extraction solution, including CI/CD integration and content filters and blocklists.
• Select the appropriate service for a knowledge mining solution, including container deployment and prompt shields with harm detection.
• Plan for a solution that meets Responsible AI principles, including resource monitoring and governance framework design.
• Create an Azure AI resource and manage cost for Microsoft Foundry Services.

Executive Summary

• Microsoft Foundry is the unified portal, resource type, and SDK surface that aggregates Azure OpenAI in Foundry Models, Azure Vision in Foundry Tools, Azure Speech in Foundry Tools, Azure Document Intelligence in Foundry Tools, Azure Language in Foundry Tools, Azure Translator in Foundry Tools, and Azure Content Understanding in Foundry Tools under a single hub and project model.
• Service selection follows three axes: capability fit, deployment SKU (Standard, Global Standard, Provisioned PTU, Batch, Data Zone Standard or Provisioned), and identity model (key, Microsoft Entra ID, managed identity).
• Responsible AI is enforced by Azure AI Content Safety, prompt shields, protected material detection, groundedness detection, content filters, blocklists, and abuse monitoring, layered on top of role based access control and data zone residency.
• Operations rely on Azure Monitor metrics, diagnostic settings, Azure Log Analytics, Azure Cost Management, and CI/CD pipelines built with Bicep, Terraform, or Azure DevOps and GitHub Actions targeting Microsoft Foundry resource definitions.

Assumptions

• All examples use Microsoft Entra ID tenants with Bicep, Azure CLI version 2.65+, Python SDKs from the azure-ai-* family, and the openai Python client targeting Azure endpoints.
• Resource naming follows the convention org-env-service-region, for example contoso-prod-aifoundry-eus2.
• Identity discipline favors managed identity over keys; keys are used only where explicitly justified.
• Region examples reference eastus2, westeurope, and swedencentral unless a specific feature requires a different region.