Skip to content

Plan, Create and Deploy an Azure AI Foundry Service

This section of the Microsoft AI-102: Designing and Implementing a Microsoft Azure AI Solution exam covers planning, creating, and deploying Azure AI Foundry services. Below are study notes for each sub-topic, with links to Microsoft documentation, exam tips, and key facts

Plan for a Solution that Meets Responsible AI Principles

📖 Docs: What is Responsible AI?

Overview

  • Responsible AI ensures AI systems are ethical, fair, and reliable
  • Microsoft defines 6 core principles:
    • Fairness
    • Reliability and safety
    • Privacy and security
    • Inclusiveness
    • Transparency
    • Accountability

Key Points

  • Must comply with data governance and regulatory standards
  • Azure provides tools such as content filters and Azure AI Content Safety
  • Responsible AI checks should be built into design, training, and deployment phases

Exam Tip

Be prepared for scenario questions on choosing controls to mitigate bias or risks

Create an Azure AI Resource

📖 Docs: Quickstart: Create your first AI Foundry resource

Overview

  • AI resources are provisioned in the Azure portal or via CLI/ARM templates
  • Services like Azure OpenAI, Vision, Language, or Speech require dedicated resources
  • Resource creation includes selecting region, pricing tier, and enabling authentication (keys or Azure AD)

Key Points

  • Some services have regional restrictions (e.g., Azure OpenAI)
  • Quotas apply per subscription and region
  • Access controlled with RBAC

Limits

Creating AI resources often requires quota approval from Microsoft

Choose the Appropriate AI Models for Your Solution

📖 Docs: Azure AI Foundry model catalog

Overview

  • The model catalog provides foundation models for text, vision, speech, and embeddings
  • Choice depends on task:
    • GPT series for text, summarization, code
    • Phi/Mistral for lightweight generative tasks
    • Embeddings for semantic search
    • Vision models for image classification/detection

Key Points

  • Consider latency, cost, accuracy, and context length
  • Prebuilt models reduce training effort, while fine-tuning allows customization

Exam Tip

Expect questions comparing fine-tuning vs RAG vs prebuilt models

Deploy AI Models Using the Appropriate Deployment Options

📖 Docs: Deployment overview for Azure AI Foundry Models

Overview

  • Deployment options include:
    • Managed endpoints (real-time inference, batch)
    • Serverless APIs (low-code usage)
    • Containerized deployments (custom runtime environments)

Key Points

  • Managed endpoints provide scaling and monitoring
  • Batch endpoints are suitable for asynchronous tasks
  • Container deployments support offline or hybrid scenarios

Use Case

  • Real-time chatbot → managed endpoint
  • Offline document processing → container deployment

Install and Utilize the Appropriate SDKs and APIs

📖 Docs: Azure AI Foundry SDK client libraries

Overview

  • SDKs provide programmatic access to AI services (Python, C#, Java, JavaScript)
  • REST APIs are available for all services
  • Common tasks:
    • Submitting inference requests
    • Managing deployments
    • Retrieving evaluation metrics

Key Points

  • SDKs simplify authentication, retries, and error handling
  • APIs are versioned and may have feature restrictions per region

Exam Tip

Memorize which SDK applies to which service (e.g., azure-ai-language, azure-ai-vision)

Determine a Default Endpoint for a Service

📖 Docs: Endpoints for Azure AI Foundry Models

Overview

  • Each Azure AI resource has a default endpoint
  • Endpoint = base URL used for API calls
  • Example format: https://<resource-name>.services.ai.azure.com/models
  • Example format: https://<resource-name>.openai.azure.com

Key Points

  • Keys or Azure AD tokens required for authentication
  • Default endpoints can be overridden by custom deployments

Best Practices

Use Azure Key Vault to manage keys and avoid hardcoding credentials

Integrate Azure AI Foundry Services into a CI/CD Pipeline

📖 Docs: CI/CD for Azure AI Foundry "AI Agent Service" Agents

Overview

  • CI/CD automates testing, deployment, and monitoring of AI services
  • Tools:
    • GitHub Actions
    • Azure DevOps Pipelines
    • ARM/Bicep templates for infra-as-code

Key Points

  • Enables version control for models and prompts
  • Supports automated testing and rollback
  • Integration with monitoring tools ensures reliability

Use Case

Auto-deploy updated prompt flows after approval in a GitHub repo

Plan and Implement a Container Deployment

📖 Docs: What are Azure AI containers?

Overview

  • Azure AI services can run in Docker containers for edge, hybrid, or disconnected environments
  • Supports services like Vision, Language, Speech, and Document Intelligence

Key Points

  • Container deployments require billing info to connect back to Azure
  • Useful for compliance (data stays local)
  • Can integrate with AKS or other orchestrators

Limits

Not all services support container deployment

Quick‑fire revision sheet

  • 📌 Responsible AI = fairness, reliability, privacy, inclusiveness, transparency, accountability
  • 📌 AI resources created in portal/CLI, with quotas and region limits
  • 📌 Choose models from model catalog based on use case
  • 📌 Deployment: managed endpoints, batch, or containers
  • 📌 SDKs available in Python, C#, Java, JS
  • 📌 CI/CD supported via GitHub Actions and Azure DevOps
  • 📌 Containers useful for hybrid or disconnected scenarios