book

Microsoft Foundry in Action

Name: Microsoft Foundry in Action
Author: Eduardo Sojo
ISBN: 9781835888681

by Eduardo Sojo

June 2026

Intermediate

370 pages

8h 22m

English

Packt Publishing

Read now

Unlock full access

Preface
Free benefits with your book
Part 1: Getting Started with Microsoft Foundry
Chapter 1: Introduction to Microsoft Foundry
Technical requirementsGetting an overview of Microsoft FoundryThe architectureRoles in Microsoft FoundryOther related Azure servicesDeploying Microsoft FoundryGetting started with using Microsoft FoundrySummaryQuestionnaire
Chapter 2: Delving Deeper into Microsoft Foundry's Features
Technical requirementsExploring Projects in the portalDiscoverModelsToolsSolution templatesBuildAgentsWorkflowsModelsFine-tuningToolsKnowledgeDataEvaluationsGuardrailsThe scenario: Building an AI support assistantOperateAssetsComplianceQuotaAdminLooking at the cost considerationsUnderstand costs before developing AI applicationsSummaryQuestionnaire
Part 2: Building with AI Models
Chapter 3: Working with Data
Technical requirements:Getting data into Microsoft FoundryFiles, folders, and tablesAzure AI SearchModel Context ProtocolExploring the RAG patternReal-world example and cost implicationsWithout RAG (Traditional approach)With RAG in Microsoft FoundryWhen to use RAG vs.Traditional APIsLooking at some other AI architecture patternsAdding data to the projectPreparing the sample datasetDeploying a modelConfiguring AI SearchUsing indexed dataSummaryQuestionnaire
Chapter 4: Choosing the Right Model
Technical requirementsChoosing a modelModel differences and data governance implicationsModel architecture differencesContext window and token limitsTraining data and knowledge cutoffOutput determinism and reproducibilityFine-Tuning capabilities and data ownershipSafety and moderation guardrailsRegulatory and compliance alignmentHow to address these governance challenges in Microsoft FoundryComparing before implementationLatency vs performance trade-offsThe core trade-offMaking the trade-off decisionFine-tuning vs.prompt engineering: Which should you choose?Prompt engineeringFine-tuningHow to decideImplementing gpt-5.2-chatDeploying the modelConfigurations for agents and modelsSummaryQuestionnaire
Chapter 5: Working with Workflows
Technical requirementsAI interaction design in Microsoft FoundryWhere Workflows fit in the portalPractical sampleUnderstanding workflowsBlank workflowSequential workflowHuman-in-the-Loop workflowGroup chat workflowChoosing the right workflow typeCreating your first workflowStep 1 – Create the workflowStep 2 – Collect the support ticketStep 3 – Invoke an agent for classificationStep 4 – Store the agent responseStep 5 – End the first versionDesigning structured and reliable outputsStep 1 – Make the agent return structured JSONStep 2 – Extract individual fieldsStep 3 – Prepare for conditional logicAdding decision logic with If/Else and routingStep 1 – Add the If/Else nodeStep 2 – Build the high-risk branch (Escalation path)Step 3 – Build the Low-Risk branch (Automated path)Connecting the workflow to an agentUnderstanding the relationship: Workflow vs.agentStep 1 – Ensure the agent is properly configuredStep 2 – Position the workflow as the entry logicStep 3 – Prepare the workflow for real useStep 4 – Think about governance and boundariesWhat we've achievedSummaryQuestionnaire
Chapter 6: Deploying AI Solutions
Technical requirementsUnderstanding deployment in Microsoft FoundryWhat can you deploy?Where deployment happens in the portalUnderstanding endpoints and consumptionDeployment as part of the AI lifecyclePreparing for deploymentValidating your core componentsTesting before deploymentChecking environment and access readinessDeployment workflow in Microsoft FoundryDeploying models from the Foundry portalNavigating to model deploymentCompleting the deploymentChoosing the right deployment typeUnderstanding edge cases in deployment selectionEdge case 1: Serverless cold starts break your SLAEdge case 2: You need both real-time AND batch processingEdge case 3: Compliance and data sensitivity eliminates choicesEdge case 4: Low RPS but extreme latency sensitivityEdge case 5: Migration from prototype to productionReal-world cost comparisonsScenario 1: Customer support chatbot (Interactive, medium volume)Scenario 2: Nightly batch analysis (Non-interactive, large workload)Deploying workflows as endpointsDeploying a workflow instead of a modelNavigating to workflow deploymentReal-world use caseCommon deployment scenarios and troubleshootingManaging deployed versionsCreating and deploying agentsUnderstanding the role of agentsCreating an agent in the Foundry portalDeploying the agentSummaryQuestionnaire
Chapter 7: Evaluating and Monitoring AI Applications
Technical requirementsFrom "It Works" to "It's Reliable"AI evaluation as a journeyAI as a living systemEvaluation in Microsoft Foundry ProjectsWhere evaluation lives in the new portalTypes of evaluationLinking evaluations to deploymentsPractical example: Compare two customer support deploymentsStep 1 – Prepare a small evaluation datasetStep 2 – Open the evaluation workspaceStep 3 – Choose what you are evaluatingStep 4 – Select your datasetStep 5 – Choose evaluation metricsReviewing results and comparing deploymentsMonitoring AI applications in productionOperational metricsConfiguring monitoring and alert settingsEvaluation metricsConfiguring continuous and scheduled evaluationsScheduled red teaming runsScheduling a red team runSummaryQuestionnaire

Part 3: Responsible AI
Chapter 8: Implementing Responsible AI Principles
Technical requirementsUnderstanding Responsible AICore concepts of Responsible AIFairnessReliability and safetyPrivacy and securityTransparency and explainabilityAccountabilityWhy Responsible AI mattersResponsible AI in Microsoft FoundryBuilt-in evaluationsGuardrails for enforcing boundariesMonitoring and observabilityHuman oversight and control pointsBringing it all togetherDesigning Responsible AI workflowsStart with intent, not promptsLayer prompts instead of relying on a single oneDesign for refusal and recoveryKeep humans in the loop where it mattersDocument assumptions and design for changeOperationalizing Responsible AIFrom design-time controls to runtime realityMonitoring signals that reflect real risksClear ownership and escalation pathsStaying audit- and compliance-readyManaging change responsiblyPutting it all together – A practical implementation walkthroughStep 1: Define scope boundaries in the system prompt and guardrailsStep 2: Build an evaluation dataset that reflects real risksStep 3: Configure monitoring for behavioral signalsStep 4: Define ownership and escalation pathsStep 5: Manage the first change responsiblyResponsible AI as an operational habitCommon pitfalls and Anti-PatternsTreating Responsible AI as a final checkboxAssuming policies automatically enforce behaviorDesigning without a clear escalation pathMistaking accuracy for trustworthinessSummaryQuestionnaire
Chapter 9: Ensuring Content Safety and Compliance
Technical requirementsUnderstanding AI safety and complianceWhy AI safety mattersThe chatbot that created legal liabilityWhat compliance means in an AI contextContent safetyGuardrailsBlocklistsDealing with false positives in content filtersBuilding continuous observability and feedback loopsComplianceBest practices for safe and compliant AI systemsDesign with safety by defaultMinimize data exposureIterate with human oversightLog transparently and responsiblyVersion everythingMonitor continuously after deploymentDay-to-day: Automated signal collectionWeekly: Pattern reviewEstablish an incident response planBringing it all togetherThe problem that surfaced in productionHow each best practice was appliedWhat made the differenceSummaryQuestionnaire
Part 4: Use Cases
Chapter 10: Building Assistants and Copilots
Technical requirementsDefining the scope of the use caseBusiness scenarioWhat the assistant will doBoundaries and data scopeSuccess criteriaCreating the basicsDeploy a language modelAdd team members to the projectPreparing the source documentsCreate a Foundry IQ knowledge baseAdd files as a knowledge sourceVerify the indexed contentConfigure knowledge base access restrictionsCreating the agentModel configurationConnect the knowledge baseAdding tool instructionsConfigure content safety filtersVerify the agent in the playgroundAdding guardrails and escalation logicCreate a custom guardrailPrompt injection, groundedness, and sensitive data detectionAssign the guardrail to the agentConfigure escalation triggers via blocklistAttach the blocklist to the guardrailTesting the assistant before it goes liveCreate the evaluation dataset fileCreate the evaluation runReview evaluation resultsDeploying and monitoring in productionPublish the agentReassign RBAC to the agent identityEnable tracing and monitoringApprove the agent in the Microsoft 365 Admin CenterOperating the system over timeScaling the agentic systemCost optimizationManaging the knowledge base over timeSummary
Chapter 11: Creating Custom Copilots with Microsoft Fabric
Technical requirementsScope and assumptionsIn-scope scenariosOut-of-scope scenariosIdentity and authentication assumptionsData exposure principlesArchitecture overviewCreating the basicsDeploy a language modelAdd team members to the projectMicrosoft Fabric as a data sourceConfirm prerequisites on the Microsoft Fabric sideCreate a Microsoft Fabric connection in FoundryAdd the Fabric data agent as a knowledge sourceVerify the Fabric data agent connectionReview the Fabric data agent schemaBuilding the agentCreate the agentSelect the model and system promptVerify the Fabric knowledge sourceAdd knowledge source routing instructionsConfigure content safety filtersVerify the agent in the playgroundAdding guardrails and compliance controlsCreate a custom guardrail and content harm controlsAdd prompt injection control and sensitive data controlAdd groundedness control and task adherence controlsCreate a compliance blocklistAssign the guardrail to the agentAttach the blocklist to the guardrailTesting the assistantCreate the evaluation dataset fileCreate the evaluation runDeploying and monitoring in productionPublish the agentReassign access to the agent identityEnable tracing and monitoringApprove the agent in Microsoft 365 Admin CenterPerformance at scaleReducing latency under customer trafficManaging schema changes without breaking the agentCompliance at scaleSummary
Chapter 12: Building an Agent that Uses Databricks Genie via MCP
Technical requirementsScope and assumptions of the use caseIn-scope scenariosRequired inputs and identity assumptionsData exposure principlesSuccess criteriaCreating the basicsDeploy a language modelAdd team members to projectConnecting the Databricks Genie via MCPConfirm prerequisites on the Databricks sideConfigure the MCP connectionOAuth consent flowVerify the Genie tool connectionReview the exposed Genie tool schemaBuilding the agentCreate the agentModel configurationConnect the Databricks genie toolAdding tool instructionsVerify the agent in the playgroundAdvanced deployment considerationsScaling and rate limitingMulti-region deploymentDisaster recovery and fallbackOperational runbooksTesting the assistantCreate the evaluation dataset fileCreate the evaluation runReview evaluation resultsDeploying and monitoring in productionPublish the agentReassign RBAC for the Databricks MCP connectionEnable tracing and monitoringOperational challenges and degradation patterns in productionRetry strategies and transient failuresSilent degradation patternsThe operational postureApproving the agent in Microsoft 365 Admin CenterSummary
Chapter 13: Unlock Your Exclusive Benefits
Unlock this Book's Free Benefits in 3 Easy Steps
Other Books You May Enjoy
Index

Overview

Build, deploy, and monitor AI applications using Microsoft Foundry's unified portal, combining models, workflows, agents, and responsible AI practices into a single development experience.

Key Features

Build, evaluate, and deploy AI solutions using Microsoft Foundry’s unified portal
Implement guardrails, evaluations, and monitoring for production-ready AI systems
Includes real-world use cases integrating Azure OpenAI and Databricks
Purchase of the print or Kindle book includes a free PDF eBook

Book Description

Unlock the full potential of AI with Microsoft Foundry, Microsoft’s unified platform for building, orchestrating, and operating AI solutions at scale. This hands-on guide walks you through the full AI application lifecycle, from data preparation and model selection to deployment, monitoring, and continuous evaluation.

Written by Eduardo Sojo, a former Microsoft consultant and current solutions architect at Databricks with over two decades of experience, the book focuses on practical implementation using the latest Microsoft Foundry portal. You’ll learn to design intelligent workflows, build agents, and apply guardrails to ensure safe, reliable AI behavior in real-world scenarios.

Rather than focusing only on models, the book shows how to connect workflows, agents, evaluations, and observability so solutions are not just functional, but production-ready. You’ll explore real-world use cases such as building copilots, integrating external systems like Databricks, and creating multi-agent architectures that work with enterprise data. By the end, you’ll be able to design, build, and operate secure, scalable, intelligent AI solutions using Microsoft Foundry.

What you will learn

Select and deploy models for vision, language, and custom AI scenarios
Use Prompt Flow to design and refine complex AI interactions
Evaluate models using standard performance metrics
Build assistants, copilots, and image-processing solutions with Azure OpenAI
Design, deploy, and orchestrate multi-agent systems using Microsoft Foundry Agent Service
Utilize Microsoft Foundry Observability for real-time monitoring and evaluation
Apply Microsoft's Responsible AI principles to real applications

Who this book is for

This book is for developers, data professionals, and AI practitioners who want to build and deploy intelligent applications using Microsoft Foundry. Whether you're transitioning from Azure AI Studio or starting fresh, this guide will help you understand how to design real-world AI systems using modern tools and best practices. Basic knowledge of programming and data concepts will help you get the most out of this book.

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.

Read now

Unlock full access

More than 5,000 organizations count on O’Reilly

O’Reilly covers everything we've got, with content to help us build a world-class technology community, upgrade the capabilities and competencies of our teams, and improve overall team performance as well as their engagement.

Julian F.

Head of Cybersecurity

I wanted to learn C and C++, but it didn't click for me until I picked up an O'Reilly book. When I went on the O’Reilly platform, I was astonished to find all the books there, plus live events and sandboxes so you could play around with the technology.

Addison B.

Field Engineer

I’ve been on the O’Reilly platform for more than eight years. I use a couple of learning platforms, but I'm on O'Reilly more than anybody else. When you're there, you start learning. I'm never disappointed.

Amir M.

Data Platform Tech Lead

I'm always learning. So when I got on to O'Reilly, I was like a kid in a candy store. There are playlists. There are answers. There's on-demand training. It's worth its weight in gold, in terms of what it allows me to do.

Mark W.

Embedded Software Engineer

Publisher Resources

ISBN: 9781835888681

Cloud Computing

Data Engineering

Data Science

AI & ML

Programming Languages

Software Architecture

IT/Ops

Security

Design

Business

Soft Skills