Design and implement a GenAIOps infrastructure
Free Microsoft MLOps Engineer Associate (AI-300) practice — 6 questions on Design and implement a GenAIOps infrastructure, with explanations. No sign-up. Full 12-question mixed test →
A company is deploying a GPT-4o model in Microsoft Foundry for a customer-facing chatbot that must handle 50,000 requests per hour with guaranteed sub-second latency during business hours (8 hours daily). The workload drops to near-zero overnight. Cost optimization is critical. Which deployment configuration BEST meets these requirements?
Managed compute with PTUs provides guaranteed throughput and consistent latency required for production SLAs. 300 PTUs can handle approximately 50,000 requests/hour for GPT-4o. Since the workload is predictable (business hours only), using Azure Automation to provision PTUs only during the 8-hour window optimizes costs by avoiding charges during idle periods while maintaining performance guarantees when needed.
An MLOps team manages 15 different prompt variants for a RAG application in Microsoft Foundry. The team needs to track prompt changes across development, staging, and production environments, enable rollback to previous prompt versions, and trigger automated evaluation pipelines when prompts are updated. The solution must integrate with existing Azure DevOps CI/CD workflows. Which approach BEST implements prompt version control?
Git repositories provide industry-standard version control with full history, branching, merging, and rollback capabilities. Storing prompts as JSON/text files allows code review through pull requests. Azure DevOps has native Git integration and can trigger pipelines on PR events, enabling automated prompt evaluation before promotion. This approach follows GitOps principles and integrates seamlessly with existing CI/CD workflows for infrastructure and model deployment.
A financial services company is deploying a Microsoft Foundry project that processes sensitive customer data using GPT-4 models. Compliance requires that all model inference traffic remain within the company's Azure virtual network and never traverse the public internet. The solution must support both prompt flow execution and direct REST API calls to deployed models. Which network configuration satisfies these requirements?
Private endpoints create private IP addresses for Foundry resources within the customer's VNet, ensuring all traffic uses private connectivity. Deploying models to managed compute with private endpoints keeps inference traffic on the private network. Virtual network integration for compute instances ensures prompt flow execution also uses private connectivity. This configuration eliminates all public internet exposure for both management and inference traffic.
An organization uses Bicep templates to deploy Microsoft Foundry infrastructure across 8 Azure regions for a global application. Each deployment includes a Foundry workspace, managed identity, Key Vault, Storage Account, and RBAC role assignments. The team needs to ensure consistent configuration while allowing region-specific parameters like location and naming prefixes. After deployment, automated tests verify that deployed resources match security baselines. Which IaC pattern BEST implements this architecture?
A parameterized main.bicep module encapsulates the complete Foundry infrastructure as reusable code. Region-specific files consume this module with regional parameters, ensuring consistency while allowing customization. The module pattern promotes DRY principles. Using --confirm-with-what-if provides deployment preview. Post-deployment Azure CLI queries (az resource show, az role assignment list) enable automated verification of deployed configuration against baselines. This approach scales efficiently across regions.
A data science team develops prompts for a Foundry-based document summarization system. They need to compare the performance of 5 prompt variants using a test set of 200 documents. The evaluation must measure summary accuracy using GPT-4 as a judge, calculate ROUGE scores, and track token usage per variant. Results must be stored for later analysis and visualized in a dashboard. Which Microsoft Foundry capability BEST supports this workflow?
Microsoft Foundry prompt flows have native variant support, allowing multiple prompt versions to be defined within a single flow. Evaluation runs can execute all variants against the same test data, compare performance using built-in metrics (ROUGE, BLEU) and custom metrics (GPT-4 as judge), and track token consumption. Results are automatically stored and displayed in Foundry's evaluation dashboard with side-by-side comparison. This is the purpose-built Foundry feature for prompt experimentation.
A company deploys 12 Microsoft Foundry projects across development, test, and production environments. Each project accesses Azure OpenAI Service, Azure AI Search, and Cosmos DB. The security team requires that compute resources in each project use managed identities (not keys or connection strings) with least-privilege RBAC assignments. Developers need permissions to create new projects but must not have access to production secrets or the ability to assign privileged roles. Which identity and access configuration BEST meets these requirements?
User-assigned managed identities can be created independently and assigned to multiple compute resources within a project, simplifying identity management. Granting service-specific data plane roles (Cognitive Services OpenAI User, Search Index Data Reader, Cosmos DB Data Reader) implements least privilege. Contributor on resource groups allows developers to create projects and resources, while deny assignments explicitly prevent any access to production resources regardless of inherited permissions. This satisfies all requirements: managed identities, least privilege, and environment isolation.
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.
Get my $34.99 deal →