Generative AI on Microsoft Azure From Large Language Models to Advanced Multi-Agent Systems (Adrián González Sánchez, Jaime De Mora etc.)（Z-Library）

Page 1

Generative AI on Microsoft Azure From Large Language Models to Advanced Multi-Agent Systems Adrián González Sánchez, Jaime De Mora & Jorge García Ximénez

Page 2

9 7 9 8 3 4 1 6 2 3 2 8 6 5 6 9 9 9 US $69.99 CAN $87.99 CLOUD COMPUTING ISBN: 979-8-341-62328-6 “As a Microsoft MVP, developer, and Azure architect, I know firsthand how critical it is to bridge the gap between AI experimentation and production-ready systems. This book does exactly what it describes— delivering expert-level depth grounded in real-world practice on the Azure ecosystem. This book is the great handbook every AI professional needs.” Jonah Andersson, Microsoft MVP and Certified Trainer and author of Learning Microsoft Azure Companies are moving generative AI projects out of the lab and into production environments. To support these sophisticated applications, they’re turning to advanced practices such as multi-agent architectures and complex code-based frameworks. This practical handbook shows you how to leverage cutting-edge techniques using Microsoft’s ecosystem of tools to deploy trustworthy AI systems tailored to your organization’s needs. Written for and by AI professionals, Generative AI on Microsoft Azure goes beyond the core aspects, examining underlying principles, tools, and practices in depth, from the art of prompt engineering to strategies for fine-tuning models to advanced techniques like retrieval-augmented generation (RAG) and agentic AI. Through real-world case studies and insights from top experts, you’ll learn how to harness AI’s full potential on Azure, paving the way for groundbreaking solutions and sustainable success in today’s AI-driven landscape. • Understand the technical foundations of generative AI and how the technology has evolved over the last few years • Implement advanced GenAI applications using services like Microsoft Foundry or Copilot, among others • Leverage patterns, tools, frameworks, and platforms to customize AI projects • Manage, govern, and secure your AI-enabled systems with responsible AI practices • Learn to avoid common pitfalls, future-proof your applications, and more Adrián González Sánchez is a Senior AI Product Manager at Microsoft AI and academic director at IE University. He’s authored several books and instructed courses for O’Reilly, LinkedIn Learning, and DeepLearning.AI. He also collaborates with 2U GetSmarter for MIT Sloan’s AI executive classes. Jaime De Mora is CTO for Digital Natives at Microsoft in EMEA, leading AI strategy and execution for startups. He designs and scales production-grade AI and multi-agent systems. He’s also a professor at IE University, focused on AI systems. Jorge García Ximénez is an AI Senior Cloud Solution Architect at Microsoft. Working at the intersection of Azure AI engineering and product strategy, Jorge brings a practical perspective on moving from experimentation to enterprise-grade AI deployment on Azure. Generative AI on Microsoft Azure

Page 3

Adrián González Sánchez, Jaime De Mora, and Jorge García Ximénez Generative AI on Microsoft Azure From Large Language Models to Advanced Multi-Agent Systems

Page 4

979-8-341-62328-6 [LSI] Generative AI on Microsoft Azure by Adrián González Sánchez, Jaime De Mora, and Jorge García Ximénez Copyright © 2026 Adrián González Sánchez, Jaime De Mora, and Jorge García Ximénez. All rights reserved. Published by O’Reilly Media, Inc., 141 Stony Circle, Suite 195, Santa Rosa, CA 95401. O’Reilly books may be purchased for educational, business, or sales promotional use. Online editions are also available for most titles (http://oreilly.com). For more information, contact our corporate/institutional sales department: 800-998-9938 or corporate@oreilly.com. Acquisitions Editor: Megan Laddusaw Development Editor: Sara Hunter Production Editor: Gregory Hyman Copyeditor: nSight, Inc. Proofreader: Krsta Technology Solutions Indexer: Potomac Indexing, LLC Cover Designer: Susan Brown Cover Illustrator: José Marzan Jr. Interior Designer: David Futato Interior Illustrator: Kate Dullea April 2026: First Edition Revision History for the First Edition 2026-04-09: First Release See http://oreilly.com/catalog/errata.csp?isbn=9798341623286 for release details. The O’Reilly logo is a registered trademark of O’Reilly Media, Inc. Generative AI on Microsoft Azure, the cover image, and related trade dress are trademarks of O’Reilly Media, Inc. The views expressed in this work are those of the authors and do not represent the publisher’s views. While the publisher and the authors have used good faith efforts to ensure that the information and instructions contained in this work are accurate, the publisher and the authors disclaim all responsibility for errors or omissions, including without limitation responsibility for damages resulting from the use of or reliance on this work. Use of the information and instructions contained in this work is at your own risk. If any code samples or other technology this work contains or describes is subject to open source licenses or the intellectual property rights of others, it is your responsibility to ensure that your use thereof complies with such licenses and/or rights.

Page 5

Table of Contents Preface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix 1. Technical Fundamentals of Generative AI Models and Applications. . . . . . . . . . . . . . . . . 1 Evolution of the Adoption Patterns 3 Types of Generative AI Models 8 Technical Concepts Behind Generative AI 10 Pretraining 11 Post-Training 16 Inference 20 Mapping to Microsoft Azure 24 Conclusion 26 2. From Prototype to Production: AI on Azure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Overview of the Advanced AI Ecosystem on Azure 27 Deep Dive on Microsoft Foundry: Models, Workflows, and Safety 30 Key Capabilities of Microsoft Foundry 30 Model Diversity and Multiprovider Strategies 35 Azure AI Model Catalog: Many Models, One Platform 35 Model Providers 36 Deployment Patterns on Azure 37 Building Agents in Azure 39 Microsoft Copilot and Copilot Studio 39 Integration Patterns with Copilot Studio 42 Putting It Together: Copilot Integration Scenarios 44 Vector Databases 45 Azure AI Search 46 Azure Cosmos DB with Vector Search 48 Selection Criteria and Best Practices 52 iii

Page 6

Typical Use Cases for Vector DBs in Generative AI Solutions 54 Additional Data Platforms 55 Azure Databricks 55 Snowflake Cortex on Azure 56 NVIDIA NIM on Azure 57 Web Application Frameworks for AI 58 Azure App Service for AI Web Apps 58 Streamlit: Rapid UI for Data and AI 59 GitHub Copilot Models and Integrations 66 Models Behind GitHub Copilot 66 Copilot Integrations 68 GitHub Copilot Integrations and Ecosystem 68 Conclusion 71 3. Model Selection and Prompt Engineering Best Practices. . . . . . . . . . . . . . . . . . . . . . . . . 73 Considerations for Model Selection 73 Choosing Your Model on the Microsoft Foundry 77 Selection via Model Catalog 77 Model Deployment Types 80 Azure OpenAI Models 82 Additional Microsoft Foundry Capabilities 85 Prompt Engineering Best Practices 87 Laying the Foundations for Effective Prompts 87 Prompt Engineering Techniques 89 Key Takeaways 91 Prompt Engineering in Microsoft Foundry 92 Conclusion 94 4. Retrieval-Augmented Generation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 Fundamentals of RAG 97 The RAG Process 98 RAG in Azure 100 RAG with Azure AI Search 103 Data Ingestion and Indexing 103 Retrieval and Generation Workflow 107 Iterative Optimization of Your RAG Pipelines 110 New Foundry and Foundry IQ 113 Advanced RAG Scenarios 115 Multimodal RAG 115 GraphRAG 115 NL2SQL 116 Agentic RAG 117 iv | Table of Contents

Page 7

RAFT 118 Looking Ahead with RAG 118 Conclusion 119 5. Fine-Tuning Generative AI Models in Azure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 When to Fine-Tune Your Generative AI Models 122 Prompt Engineering Versus RAG Versus Fine-Tuning 123 Fine-Tuning Taxonomy 124 SFT 124 PEFT 125 Fine-Tuning in Microsoft Foundry 126 Fine-Tuning Azure OpenAI Models 128 Step-by-Step Fine-Tuning Workflow for Azure OpenAI Models 130 Step-by-Step Workflow for Fine-Tuning Other Catalog Models in Microsoft Foundry 135 Fine-Tuning Best Practices and Pitfalls 138 Conclusion 140 6. Agentic Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 When to Use Agentic Systems 145 Agentic Protocols 147 MCP in Detail 148 Agentic Design Patterns 152 Sequential Pattern 152 Workflow Pattern 153 Concurrent Pattern 153 Group-Chat Pattern 154 Self-Reflection/Cross-Reflection Pattern 155 Handoff Pattern/Human in the Loop 156 Supervisor Pattern 157 Magentic Orchestration Pattern 157 Summary of Patterns 159 Building Agentic Systems in Azure 159 How the Microsoft Foundry Agent Service Works 159 Supercharging Agents: Tool Catalog and Foundry IQ 161 Foundry Workflows: Orchestrating Multiagent Processes 162 Microsoft Agent Framework: From Semantic Kernel and AutoGen to Present 165 Hosted Agents: Deploying Agent Framework Apps to Azure 166 Conclusion 166 Table of Contents | v

Page 8

7. GenAIOps and LLMOps in Azure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 From MLOps to GenAIOps: Why It Matters 169 Generative AI in Azure: The Operating Loop 171 Use Case: Multiagent Financial Compliance Assistant 172 Define and Explore 173 Quantitative Analysis and Monitoring: Core Model Performance Metrics 173 Model Selection and Evaluation 174 Tools and Frameworks 175 Build and Customize 177 Designing the Application Architecture 177 Build and Customize: Example 178 Experimentation and Prompt Engineering 179 Evaluation and Validation 180 Code Templates and Accelerators 182 Fine-Tuning Guidance 183 Observe and Optimize 184 CI/CD 184 Deployment and Release Management 187 Monitoring in Production 188 Automation of Model Lifecycle Management 191 Tools for Tracking, Versioning, and Reproducibility 192 Protect and Govern 193 Safety and Guardrails by Design 193 Security and Access Control 196 Governance and Accountability 198 Linking Protect and Govern with the GenAIOps Lifecycle 199 Conclusion 200 8. Generative AI Governance Framework. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 Thought Process for Your AI Governance 201 Levels of AI Governance 202 Typical AI Governance Assets 203 Internal Context at Organizational Level 207 Other Considerations 208 Key Building Blocks 214 Azure AI Content Safety 214 Microsoft Purview 216 Microsoft Defender 218 Microsoft Entra 220 Microsoft Responsible AI (RAI) Toolbox 222 PyRIT (Python Risk Identification Tool) 224 HAX (Human-AI eXperience) Toolkit 226 vi | Table of Contents

Page 9

AI Bill of Materials (AI BOM) 228 HiddenLayer 230 How-To: Implementing Generative AI Governance on Azure 232 Preliminary, General First Steps 233 Human + AI Approach 234 Securing Your AI Systems Against Human Misuse 236 AI Models, Data, and Infrastructure 237 Last but Not Least: Are You the Responsible AI Champion? 238 Conclusion 239 9. Expert Interviews. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 Innovating and Exploring with Microsoft Azure, with Marco Casalaina 241 The Art of Designing AI Architectures, with James Serra 247 Building the Future of AI Platforms, with Eric Boyd 253 The Evolution and Future of Frontier AI, with Yina Arenas 256 A. AI Performance Metrics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263 B. Additional Resources and Reading Materials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267 Glossary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275 Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289 Table of Contents | vii

Page 10

(This page has no text content)

Page 11

Preface Generative AI (GenAI) is evolving at a pace few technologies have ever matched. What began as fascination with increasingly powerful models has quickly matured into something far more complex and consequential: platforms, architectures, gover‐ nance frameworks, and production systems that reshape how modern software is designed and operated. This book is written from the field, by three practitioners and friends who live this transformation every day. Each of us brings a distinct perspective shaped by hands- on work with enterprise customers, startups, and global engineering communities. Together, our voices reflect different angles of the same reality: GenAI is no longer an experiment, and Microsoft Azure is changing the way companies implement enterprise-grade solutions. Adrián: Crossing the Generative AI Chasm Here we are. A few years after the initial boom of GenAI technologies, and not so long ago from my previous book, Azure OpenAI Service for Cloud Native Applications, I’m sitting down again to write another book. And trust me, it is not to repeat topics that may look quite similar. While that book covered Azure OpenAI Service as a specific type of GenAI model, this new work explores the broader evolution of GenAI into comprehensive platforms and scalable deployments. What started as a model-centric area of innovations (because everyone used to focus on the models), with the Azure OpenAI Service being a pioneer of enterprise-grade AI models, has truly evolved into something bigger, better, and way wider than “just” GenAI models. It is about platforms and deployments at scale, supported by architectural patterns with cloud native technologies. From my work on the field as an AI architect at Microsoft, I can affirm that the industry in general is learning how to use these technologies, and companies are understanding the new level of technical, financial, and operational complexity related to this kind of project. I must admit that I still get pleasantly surprised when I ix

Page 12

see unexpected examples of great implementations from some big companies, digital native startups, and even public administrations. And this is already happening, more often than ever, with new use cases that sometimes exceed my regular knowledge and help me continue my upskilling journey. But there is still a huge gap between those top GenAI adopters and other organiza‐ tions, or, concretely, what Geoffrey A. Moore describes in his masterpiece Crossing the Chasm as a mainstream market with “late majority” and “laggards.” And part of the market is still waiting to get started or to find the magic formula to make sense and make the most of GenAI. This is even more important now that we have regulations such as the EU AI Act that require adopter companies to develop AI literacy, which implies training their employees to understand and use AI technologies. That is the first reason behind this new book: educating an audience that needs to innovate more and better but ensuring that they first learn and understand key concepts. Let’s now come back to my first point about initial model centricity related to GenAI innovations. For some time, this area has been similar to the megapixel battle in which cell phone brands had to get the biggest number before their competitors. A similar type of competition has happened with GenAI, the number of billions of parameters, and the quantitative results from industry benchmarks such as LMArena (formerly Chatbot Arena), ARC-AGI, or Codeforces, among others. But it is not about models anymore or at least not only about that. Companies are combining multiple model providers like Azure OpenAI, Mistral, Meta’s Llama, DeepSeek, etc., and leveraging platforms that go beyond core large language model (LLM) capabilities. For that reason, cloud hyperscalers like Microsoft have continued evolving their platforms (understood as the cloud ecosystems that tie together models, data, and tools for scalable AI) to add different layers of cus‐ tomization based on internal company knowledge bases, cybersecurity and content safety measures, newly automated performance monitoring metrics, and the ability to combine and orchestrate in the same application different GenAI engines (we call them “AI agents,” and you will learn more about them in this book). This is the second reason for this book: while the previous Azure OpenAI book focused on a very specific part of the Microsoft Azure AI stack and a specific series of models, this book considers the whole AI platform, including Azure and other Microsoft tools, and the wide variety of first- and third-party models and frameworks. And that’s a natural consequence given the unstoppable evolution of the multiple product roadmaps in the Microsoft ecosystem. The third reason for this new book is the impact beyond technology: new use cases and business models, but also an impressive transformation of what solutions and their interfaces look like and the way users interact with them, thanks to new AI-enabled features. As you can see in Figure P-1, this includes computer-use automation for screen-based tasks, buttons, and other user interface (UI) elements x | Preface

Page 13

that include AI capabilities to retrieve and synthesize information in just one click, or even new possibilities for adaptive interfaces that change their shape and content depending on the user context and journey. All these rely on different types of language, reasoning, and automation models, and their combination with existing and new UIs. Figure P-1. New generation of AI-enabled interfaces Exciting times to release this book, and I am even more excited to do this with two esteemed friends: Jorge García Ximénez and Jaime De Mora. Both of them are Microsoft colleagues and top professionals with great experiences and connections within the industry. This will provide a great upgrade for you and will give you the most updated guide to leverage Generative AI on Microsoft Azure directly from the field, an A-Z coverage of Microsoft Azure without any beating around the bush. Jaime: Shipping AI Responsibly in the Real World A few years ago, most conversations about AI were theoretical. They happened in research papers, conference talks, or small side projects running safely away from anything mission critical. Today, those conversations happen in boardrooms, prod‐ uct reviews, and late-night Slack/Teams threads just before a launch. The tone has changed. The questions are no longer “Is this possible?” but “Can we ship this?” “Can we trust it?” and “What happens if it goes wrong?” Artificial intelligence is transforming almost every industry at once. Adoption curves are steepening in a way I’ve rarely seen before. What took cloud computing a decade is happening in AI in a handful of years. Teams are moving from experimentation to production under intense competitive pressure, often with incomplete information, Preface | xi

Page 14

fragmented tooling, and very real risks. And yet, despite the progress, much of AI still feels like a black box. Powerful, yes, but opaque, probabilistic, and difficult to reason about when things break. I see this tension every week in my work. I spend my time with some of the fastest- growing technology companies in the world, startups and scaleups that are pushing the boundaries of what AI-powered software can do. These are exceptional teams: strong engineers, ambitious founders, deep technical talent, and access to the best models and platforms available. And still, they struggle. Not because they lack vision or capability but because implementing AI well is genuinely hard. The complexity rarely lies in calling a model. It lies in everything around it: choosing between an ever-growing set of frameworks; deciding how to connect models to data safely; designing systems that can be evaluated, monitored, and debugged when behavior changes; managing cost, latency, security, compliance, and governance, often all at once; and doing so while the business is under pressure to move fast, ship features, and avoid being disrupted by competitors who are racing just as hard. I’ve been in rooms where a demo worked perfectly, but no one felt comfortable putting it into production. I’ve seen architectures that looked elegant on paper become liabilities six months later. I’ve watched teams rewrite entire systems not because the model got worse but because the surrounding infrastructure couldn’t support growth, reliability, or accountability. These moments are rarely visible from the outside, but they define whether AI becomes a real advantage or a lingering risk. What strikes me most is that the challenges are remarkably consistent across compa‐ nies and industries. The same questions come up again and again. How do we move beyond prototypes? How do we reduce uncertainty? How do we design systems we can explain, control, and improve over time? And perhaps most importantly: how do we build AI systems we can stand behind when they are no longer experiments but part of how the business operates? This book exists because I’ve seen what happens on both sides of that divide. When teams get it wrong, AI becomes fragile, expensive, and stressful. Progress slows. Trust erodes. But when teams get it right (when architecture, tooling, governance, and responsibility are treated as first-class concerns), the upside is enormous. AI stops being a novelty and becomes leverage. It augments teams, reshapes products, and opens entirely new possibilities for scale and impact. My perspective is shaped by working at the intersection of technology, execution, and pressure. I don’t approach AI as a purely academic problem or as a hype cycle to be chased. I approach it as a systems problem, one that demands rigor, pragmatism, and humility. The goal is not to build something impressive once but to build something that continues to work when the model changes, the data grows, regulations evolve, and users depend on it. xii | Preface

Page 15

This book is not about chasing the latest model or framework. It is about cracking the harder nut: how to design, implement, and operate AI systems responsibly in the real world. This means systems that acknowledge uncertainty instead of hiding it; systems that can be evaluated, monitored, and improved; and systems that balance speed with safety and balance innovation with accountability. Together with Adrián and Jorge, we wrote this book for practitioners who feel the weight of these decisions. We also wrote it for engineers, architects, and leaders who know that AI is no longer optional but also know that getting it wrong has real consequences. If there is one message I hope you take from these pages, it is this: the value of AI does not come from how powerful the models are but from how well we design the systems around them. And that, ultimately, is a responsibility worth taking seriously. Jorge: Architecting Enterprise-Grade Generative AI When I look at where GenAI started for most organizations and where it is heading today, one thing is clear: we have moved far beyond the era of experimenting with a single model behind a chat interface. The real transformation happens when these models are integrated into systems, connected to enterprise data, adapted to specific tasks, governed responsibly, and orchestrated together to solve complex problems. In my role as a Global Cloud Solution Architect in AI Apps at Microsoft, I work daily with customers who are transitioning from proofs of concept to production- grade GenAI platforms. What consistently emerges is that success is rarely about picking the “best” model. Instead, it is about architecture: how you combine retrieval- augmented generation (RAG) with enterprise knowledge, when fine-tuning actually adds value versus operational risk; how you design for cost, latency, evaluation, and safety; and increasingly how you compose intelligent systems made of multiple collaborating AI agents. At the same time, I have the privilege of helping scale AI knowledge internally as one of the leads of Microsoft’s AI & ML Community: a global community of more than 23,000 engineers, architects, researchers, and field roles across the company. Through hundreds of deep technical sessions, skilling programs, and real-world case discussions, I’ve seen a clear pattern: organizations are hungry not just for model capabilities but for practical patterns that turn GenAI into reliable, maintainable, and impactful software systems. In this book you will explore how modern GenAI applications are built in Micro‐ soft Azure using layered architectures that combine data platforms, vector search, evaluation pipelines, observability, governance, and orchestration frameworks. You will learn when RAG is the right tool and when it becomes a bottleneck. You will see how fine-tuning fits into enterprise scenarios, not as a default choice but as a strategic Preface | xiii

Page 16

optimization. And you will dive into agentic systems, where multiple specialized models and tools collaborate to automate workflows, reason across complex tasks, and interact with real software environments. My goal with this book is simple: to help you think like a GenAI system architect. This means not just how to call a model but how to design robust platforms that can grow from a single use case into enterprise-wide AI capabilities, securely, responsibly, and sustainably. Together with Adrián and Jaime, we focus on helping readers industrialize GenAI systems by moving from isolated experiments to robust, scalable, and governed production platforms on Microsoft Azure. GenAI is no longer a feature. It is becoming a core layer of modern software. And the organizations that master its architectures will be the ones that truly unlock its value. How This Book Is Organized This book presents a modular structure that allows you to enjoy each chapter sepa‐ rately, but all chapters together follow a natural sequence of topics that will facilitate your GenAI on Microsoft Azure skilling journey. Figure P-2 shows how these book chapters will build your AI skills, from fundamentals to real-world case studies. Figure P-2. Book chapters and covered topics Chapter 1: Technical fundamentals of GenAI The first chapter is not just a regular introduction to GenAI but also a great overview of the main technical concepts related to the entire model lifecycle, including the steps and techniques that are usually far from regular users (e.g., pre- and post-training activities). Also, it serves as the initial high-level mapping of all relevant concepts to specific Microsoft-related and Microsoft Azure serv‐ ices, and it sets the level for the rest of the book with highly technical concepts, explained (hopefully!) in an accessible and easy-to-understand manner. xiv | Preface

Page 17

Chapter 2: Advanced AI with Microsoft Azure and others The second chapter builds on the initial mapping of services, and goes deeper into the full list of relevant Microsoft Azure and other resources, such as Micro‐ soft Copilot, Copilot Studio, Snowflake Cortex, Azure Databricks, NVIDIA on Azure, and GitHub Models. It also connects with important building blocks such as the endless list of LLM and AI agent frameworks, available vector databases, and web application frameworks. This chapter is certainly the most complete guide of all GenAI things on Azure and a good way for you to understand all the potential options for your AI developments. Chapters 3 to 6: Mix of model adoption topics and agentic patterns These four chapters are the perfect mix to understand the GenAI model choice criteria and the customization patterns for GenAI on Azure. The chapters cover topics from the RAG versus fine-tuning dilemma to the most recent AI agents (and multiagents) topics, including prompt engineering topics for different types of GenAI models. Chapter 7: GenAIOps and MLOps in Azure This chapter enters a very nascent topic, such as the denominated GenAIOps or LLMOps, which is the domain or discipline related to the productization and monitoring of GenAI-enabled systems. It also covers all relevant MLOps topics for non-GenAI systems. From a Microsoft Azure perspective, this part of the book highlights specific areas of the Microsoft Foundry, with relevant evaluation and performance monitoring topics that enable model validation at scale, in a programmatic way, before and after prod-level deployments. Chapter 8: GenAI governance framework When we talk about “governance,” we know that this encompasses multiple topics, including responsible AI (RAI), AI security and safety, compliance, and that mix of data and AI governance that is becoming so relevant for any GenAI adopter out there. Because of the nature of the topic, this chapter will include a mix of definitions and key concepts, along with the very specific, technical con‐ cepts related to Microsoft Azure and other Microsoft platforms such as Defender or Purview. Chapter 9: Expert interviews Because no content can replace the experience from the field, the last chapter will include real-world use cases, along with high-quality expert interviews that will complement your learning and reading experience. From adopter companies to key Microsoft experts, this is the best possible ending ever for this book. Preface | xv

Page 18

Appendices A and B and the Glossary As any other O’Reilly or similar technical books, we will include quick references to official documentation, services, architectures, and other reading materials that may complement your experience. In summary, the mix of chapter topics covers everything you need to first start then advance your GenAI experience with Microsoft Azure. This includes everything from the field, as advanced as you can get it, and as updated as this crazy pace of innovation allows it. Let’s now get started with Chapter 1! Conventions Used in This Book The following typographical conventions are used in this book: Italic Indicates new terms, URLs, email addresses, filenames, and file extensions. Constant width Used for program listings, as well as within paragraphs to refer to program elements such as variable or function names, databases, data types, environment variables, statements, and keywords. This element signifies a general note. O’Reilly Online Learning For more than 40 years, O’Reilly Media has provided technol‐ ogy and business training, knowledge, and insight to help companies succeed. Our unique network of experts and innovators share their knowledge and expertise through books, articles, and our online learning platform. O’Reilly’s online learning platform gives you on-demand access to live training courses, in-depth learning paths, interactive coding environments, and a vast collection of text and video from O’Reilly and 200+ other publishers. For more information, visit https://oreilly.com. xvi | Preface

Page 19

How to Contact Us Please address comments and questions concerning this book to the publisher: O’Reilly Media, Inc. 141 Stony Circle, Suite 195 Santa Rosa, CA 95401 800-889-8969 (in the United States or Canada) 707-827-7019 (international or local) 707-829-0104 (fax) support@oreilly.com https://oreilly.com/about/contact.html We have a web page for this book, where we list errata, examples, and any additional information. You can access this page at https://oreil.ly/genai-on-azure. For news and information about our books and courses, visit https://oreilly.com. Find us on LinkedIn: https://linkedin.com/company/oreilly. Watch us on YouTube: https://youtube.com/oreillymedia. Acknowledgments From the three authors, thank you to the amazing O’Reilly team (Sara Hunter, Megan Laddusaw, Kristen Brown, Gregory Hyman, etc.) for the effort and trust, and to the incredible set of technical reviewers (Vaibhav Gujral, Tom Taulli, Laxmi Vanam, Alexandre Levret, and Jack Lee) who helped us convert these pages into an actual complete book. Preface | xvii

Page 20

(This page has no text content)