• AI engineers building production AI applications.
• Data scientists moving models from notebooks into governed operations.
• DevOps engineers supporting AI deployment pipelines.
• Cloud architects designing scalable AI infrastructure.
• Technical managers responsible for AI governance, reliability, and cost control.
• Federal and commercial teams deploying AI into secure Azure environments.

Prerequisites

Working knowledge of:

• Basic Python experience.
• Familiarity with machine learning concepts.
• Basic Azure knowledge.
• Familiarity with Git or GitHub.
• Introductory understanding of CI/CD.
• Awareness of responsible AI, security, and cloud governance concepts.

You will Learn:

• Design secure MLOps and GenAIOps infrastructure on Azure.
• Use Azure Machine Learning workspaces, assets, compute, environments, and registries.
• Automate AI infrastructure with Bicep, Azure CLI, GitHub, and GitHub Actions.
• Train, track, register, version, deploy, and monitor machine learning models.
• Deploy and manage generative AI applications and agents using Microsoft Foundry.
• Evaluate generative AI quality, safety, groundedness, cost, and performance.
• Optimize RAG, embedding, fine-tuning, and model performance for production workloads.

Course Outline: AI-300 Operationalize Machine Learning and Generative AI Solutions

Module 1: Introduction to AI Operations, MLOps, and GenAIOps

Business benefit:
Organizations are moving beyond AI pilots into production AI systems. This module explains how AI operations help teams deliver reliable, secure, monitored, and scalable AI solutions that support real business outcomes.

Topics

• Understand the difference between experimentation and production AI.
• Define MLOps, GenAIOps, and broader AI operations.
• Identify the roles of data scientists, AI engineers, DevOps teams, and business stakeholders.
• Review the Azure services used for operational AI solutions.
• Map AI operations to governance, reliability, security, and cost management.

Lab:
Create an AI operations reference architecture for a business use case such as customer service prediction, document intelligence, or generative AI knowledge search.

Module 2: Design and Implement MLOps Infrastructure

Microsoft’s AI-300 exam objectives include designing and implementing MLOps infrastructure, creating Machine Learning workspaces, managing datastores, compute targets, identity, assets, environments, components, registries, and infrastructure as code.

Business benefit:
A well-designed MLOps foundation reduces deployment risk, improves repeatability, and allows AI teams to scale from individual experiments to enterprise-grade machine learning operations.

Topics

• Create and manage Azure Machine Learning workspaces.
• Configure datastores, compute targets, and shared assets.
• Manage identity and access with role-based access control.
• Create reusable environments, components, and data assets.
• Share assets across workspaces using registries.

Lab:
Provision an Azure Machine Learning workspace and configure compute, datastore, identity, and reusable assets for a production AI project.

Module 3: Automate AI Infrastructure with GitHub Actions, Azure CLI, and Bicep

Microsoft includes GitHub integration, Bicep, Azure CLI, automated provisioning, secure workspace access, network restrictions, and Git source control as part of AI-300’s MLOps infrastructure objectives.

Business benefit:
Infrastructure as code enables repeatable deployments, improves governance, and supports secure promotion of AI solutions across development, test, and production environments.

Topics

• Implement infrastructure as code for Azure Machine Learning.
• Use Bicep templates to deploy AI resources.
• Use Azure CLI for repeatable AI operations.
• Configure GitHub Actions for automated provisioning.
• Apply secure networking and restricted access patterns.

Lab:
Build a GitHub Actions workflow that deploys Azure Machine Learning infrastructure using Bicep and Azure CLI.

Module 4: Orchestrate Machine Learning Model Training

AI-300 includes MLflow experiment tracking, automated machine learning, notebooks, hyperparameter tuning, training scripts, distributed training, training pipelines, and model performance comparison.

Business benefit:
Model training must be traceable, repeatable, and measurable. This module teaches teams how to manage training workflows so models can be compared, improved, and approved for production.

Topics

• Track experiments using MLflow.
• Use notebooks for exploration and controlled experimentation.
• Run training scripts in Azure Machine Learning.
• Automate hyperparameter tuning.
• Build training pipelines and compare model performance.

Lab:
Train and track multiple model versions using MLflow, compare results, and select a candidate model for registration.

Module 5: Register, Version, and Govern Machine Learning Models

Microsoft’s AI-300 objectives include MLflow model registration, responsible AI evaluation, feature retrieval packaging, lifecycle management, and model archiving.

Business benefit:
Production AI requires governance. Teams need to know which model is deployed, why it was approved, how it performs, and when it should be replaced or retired.

Topics

• Register MLflow models.
• Version model artifacts and metadata.
• Package model dependencies and feature retrieval specifications.
• Evaluate models using responsible AI principles.
• Manage model lifecycle, promotion, approval, and archiving.

Lab:
Register a trained model, document model metadata, evaluate responsible AI considerations, and prepare the model for controlled deployment.

Module 6: Deploy Machine Learning Models to Production

AI-300 measures deployment of models as real-time or batch endpoints, testing and troubleshooting model endpoints, progressive rollout, and safe rollback strategies.

Business benefit:
Deployment is where AI becomes operational. This module focuses on reliable release practices so organizations can deploy models safely without disrupting business processes.

Topics

• Deploy models as managed online endpoints.
• Deploy models as batch endpoints.
• Test and troubleshoot inference endpoints.
• Implement progressive rollout strategies.
• Plan safe rollback and version recovery.

Lab:
Deploy a model to a managed endpoint, test predictions, simulate an issue, and roll back to a prior model version.

Module 7: Monitor and Maintain Machine Learning Models

Microsoft includes data drift detection, production performance monitoring, and retraining or alert triggers when thresholds are exceeded.

Business benefit:
Models degrade over time as business data changes. Monitoring helps organizations detect drift, reduce operational risk, and keep AI outcomes aligned with business expectations.

Topics

• Monitor production model performance.
• Detect and analyze data drift.
• Define thresholds for alerts and retraining.
• Review model health and operational metrics.
• Plan ongoing model maintenance and lifecycle reviews.

Lab:
Configure monitoring logic for a deployed model and design an alert and retraining workflow.

Module 8: Design and Implement GenAIOps Infrastructure with Microsoft Foundry

AI-300 includes creating and configuring Foundry resources and project environments, managed identities, RBAC, private networking, Bicep templates, and Azure CLI deployments.

Business benefit:
Generative AI applications and agents require more than prompts. They need secure environments, controlled model access, governance, observability, and deployment discipline.

Topics

• Create and configure Microsoft Foundry resources.
• Set up project environments for generative AI workloads.
• Configure managed identities and RBAC.
• Implement secure networking and private access.
• Deploy GenAIOps infrastructure using Bicep and Azure CLI.

Lab:
Create a Microsoft Foundry project environment and configure identity, access, and infrastructure settings for a production generative AI solution.

Module 9: Deploy and Manage Foundation Models for Production Workloads

Microsoft’s AI-300 GenAIOps objectives include deploying foundation models using serverless API endpoints and managed compute, selecting models for use cases, implementing model versioning, production deployment strategies, and configuring provisioned throughput units.

Business benefit:
Choosing and operating the right foundation model affects cost, latency, security, quality, and business value. This module teaches model selection and deployment strategy for enterprise workloads.

Topics

• Select appropriate foundation models for business use cases.
• Deploy models using serverless API endpoints.
• Use managed compute options where appropriate.
• Apply model versioning and production deployment strategies.
• Configure provisioned throughput for high-volume workloads.

Lab:
Compare model deployment options for a customer service, finance, or knowledge-management use case and select the best production pattern.

Module 10: Manage Prompts, Prompt Variants, and Source Control

AI-300 includes prompt design and development, creating prompt variants, comparing performance across prompts, and implementing prompt version control using Git repositories.

Business benefit:
Prompts are production assets. Managing prompt versions helps teams improve quality, reduce risk, and maintain change control for generative AI applications.

Topics

• Design prompts for enterprise use cases.
• Create prompt variants for testing and optimization.
• Compare prompt performance across evaluation criteria.
• Store prompts in Git repositories.
• Promote approved prompts through controlled environments.

Lab:
Create and version multiple prompt variants, compare output quality, and promote an approved prompt to a simulated production branch.

Module 11: Evaluate Generative AI Applications and Agents

AI-300 includes test datasets, data mapping, quality metrics such as groundedness, relevance, coherence, and fluency, risk and safety evaluation, harmful content detection, and automated evaluation workflows.

Business benefit:
Generative AI quality must be measured, not guessed. This module helps organizations establish evaluation practices that improve trust, compliance, safety, and user adoption.

Topics

• Create test datasets for generative AI evaluation.
• Map inputs, outputs, context, and expected results.
• Measure groundedness, relevance, coherence, and fluency.
• Configure safety and harmful content evaluations.
• Automate evaluation workflows with built-in and custom metrics.

Lab:
Build an evaluation dataset and score a generative AI application using quality, safety, and business-relevance criteria.

Module 12: Implement Observability for Generative AI Applications and Agents

Microsoft’s AI-300 objectives include continuous monitoring in Foundry, latency, throughput, response times, token consumption, resource usage, logging, tracing, and debugging.

Business benefit:
Observability helps teams understand how generative AI systems behave in production, manage costs, troubleshoot problems, and improve user experience.

Topics

• Monitor generative AI applications in Microsoft Foundry.
• Track latency, throughput, and response times.
• Monitor token consumption and resource usage.
• Configure logging, tracing, and debugging.
• Use observability data to improve operations and cost control.

Lab:
Configure monitoring and logging for a generative AI application and analyze performance, cost, and troubleshooting metrics.

Module 13: Optimize Retrieval-Augmented Generation Performance

AI-300 includes optimizing RAG retrieval performance with similarity thresholds, chunk sizes, retrieval strategies, embedding model selection, hybrid search, relevance metrics, and A/B testing.

Business benefit:
RAG is central to many enterprise AI solutions. Better retrieval improves answer quality, reduces hallucination, and helps AI systems use trusted organizational knowledge.

Topics

• Tune similarity thresholds.
• Optimize chunk size and retrieval strategy.
• Select embedding models for domain-specific use cases.
• Implement hybrid semantic and keyword search.
• Evaluate RAG quality using relevance metrics and A/B testing.

Lab:
Tune a RAG pipeline by adjusting chunking, retrieval, and evaluation settings to improve answer quality.

Module 14: Fine-Tune and Customize Generative AI Models

AI-300 includes advanced fine-tuning methods, synthetic data for fine-tuning, monitoring fine-tuned model performance, and managing fine-tuned models from development through production.

Business benefit:
Fine-tuning and model customization can improve domain-specific performance, but they must be governed, tested, monitored, and cost-justified.

Topics

• Identify when fine-tuning is appropriate.
• Design fine-tuning approaches for business scenarios.
• Create and manage synthetic data for fine-tuning.
• Monitor performance of fine-tuned models.
• Manage fine-tuned models through production deployment.

Lab:
Develop a fine-tuning decision plan comparing prompt engineering, RAG optimization, and fine-tuning for a production scenario.

Module 15: Capstone — Production AI Operations Solution

Business benefit:
This capstone brings together MLOps, GenAIOps, automation, monitoring, governance, and optimization into a practical enterprise implementation scenario.

Capstone scenario options

• AI-powered customer service knowledge assistant.
• Predictive maintenance model with generative AI reporting.
• Finance anomaly detection with AI-generated explanations.
• Federal case management AI assistant with secure data access.
• HR or operations knowledge assistant with monitored RAG.

Capstone activities

• Design the Azure AI operations architecture.
• Define MLOps and GenAIOps deployment pipelines.
• Configure governance, security, and access controls.
• Define monitoring, quality, and cost metrics.
• Present a production readiness plan.