Free Microsoft MLOps Engineer Associate (AI-300) practice — 6 questions on Design and implement an MLOps infrastructure, with explanations. No sign-up.
Full 12-question mixed test →
Question 1 of 6 · Design and implement an MLOps infrastructure
A company requires all Azure Machine Learning workspaces to be deployed with private endpoints and needs to automate workspace creation across multiple environments (dev, staging, production). The deployment must include compute clusters with specific VM SKUs and a managed virtual network with approved outbound rules. Which approach provides the MOST maintainable and repeatable solution?
Bicep provides Infrastructure as Code with better maintainability than ARM JSON, Azure CLI enables parameterized deployments, GitHub integration provides version control and audit trails, and GitHub Actions automates the entire provisioning pipeline. This approach supports managed VNet configuration directly in Bicep and handles private endpoint requirements declaratively.
Question 2 of 6 · Design and implement an MLOps infrastructure
An MLOps team needs to share trained model components and curated datasets across three Azure Machine Learning workspaces in different regions. The assets must maintain version lineage and the solution should minimize data transfer costs when teams in each region deploy models. What is the BEST configuration?
Azure ML registries enable cross-workspace asset sharing while maintaining version lineage and governance. Asset materialization allows data references to region-specific storage, minimizing egress costs. Components and environments in registries are immutable and versioned, supporting MLOps best practices.
Question 3 of 6 · Design and implement an MLOps infrastructure
A financial services company must restrict all network access to Azure Machine Learning workspaces to corporate IP ranges only and prohibit internet access from compute instances. Training jobs need access to Azure Key Vault for credentials and Azure Container Registry for custom environments. Which network configuration satisfies these requirements?
Disabling public access with managed VNet in 'Allow Only Approved Outbound' mode prevents internet access while permitting specific Azure services via approved rules. This configuration enforces strictest network isolation while enabling required service dependencies. Managed identity ensures secure authentication without embedded credentials.
Question 4 of 6 · Design and implement an MLOps infrastructure
An organization uses GitHub as their source control and needs to automate Azure ML workspace deployment when infrastructure code changes are merged to the main branch. The deployment workflow must authenticate using federated credentials without storing long-lived secrets. What is the correct implementation?
OIDC federation with federated credentials eliminates long-lived secrets by using short-lived tokens issued by GitHub. The azure/login action natively supports OIDC authentication. This is the secure, Microsoft-recommended approach for GitHub Actions to Azure authentication as of 2026.
Question 5 of 6 · Design and implement an MLOps infrastructure
A data science team needs to register a new version of a training dataset stored in Azure Data Lake Storage Gen2. The data includes customer information that must be tracked for lineage purposes. The dataset should be accessible across multiple pipelines using a consistent reference. The storage account uses hierarchical namespace and contains 50GB of parquet files. What is the correct approach to register this as an Azure ML data asset?
URI folder assets are the modern approach for folder-based data in Azure ML (as of 2026, File/Tabular datasets are legacy). The abfss:// protocol is correct for ADLS Gen2 hierarchical namespace. Versioning enables lineage tracking, and URI assets provide flexible, path-based references usable across pipelines without data movement.
Question 6 of 6 · Design and implement an MLOps infrastructure
An MLOps engineer needs to create a reusable Python environment for training jobs that includes TensorFlow 2.15, custom internal packages from a private Azure Artifacts feed, and must work on both CPU and GPU compute clusters. The environment should be shared across multiple workspaces via a registry. What is the MOST appropriate implementation?
Custom Docker images provide full control over environment configuration and work identically on CPU/GPU (CUDA libraries are present but unused on CPU). Using the Azure ML base image ensures compatibility. Building and pushing to ACR enables sharing via registry. Including Azure Artifacts authentication in build (not runtime) is the correct pattern for private packages. The single image approach with CUDA support works across both compute types.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$129.99$34.99 with code FREETEST33 — valid through September 2.
