Generative AI at AWS Turn business strategy into production-ready AI applications and agents (Nestor Gandara, Eduardo Ordax etc.)（Z-Library）

📄 File Format: PDF

💾 File Size: 6.3 MB

📖 Read Online ⬇️ Download

Registered users can read the full content for free

Register as a Gaohf Library member to read the complete e-book online for free and enjoy a better reading experience.

📄 Page 1

(This page has no text content)

📄 Page 2

Generative AI at AWS Turn business strategy into production-ready AI applications and agents Nestor Gandara Eduardo Ordax Srikanth Daggumalli Ashutosh Dubey

📄 Page 3

Generative AI at AWS Copyright © 2026 Packt Publishing All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form or by any means without the prior written permission of the publisher, except in the case of brief quotations embedded in critical articles or reviews. Every effort has been made in the preparation of this book to ensure the accuracy of the information presented. However, the information contained in this book is sold without warranty, either express or implied. Neither the authors nor Packt Publishing, nor its dealers and distributors, will be held liable for any damages caused or alleged to have been caused directly or indirectly by this book. Packt Publishing has endeavored to provide trademark information about all of the companies and products mentioned in this book by the appropriate use of capitals. However, Packt Publishing cannot guarantee the accuracy of this information. This book was written by Nestor Gandara, Eduardo Ordax, Srikanth Daggumalli, and Ashutosh Dubey. Generative AI tools were used only to assist with ideation, phrasing, and diagram drafts, and all technical content and code were created, verified, and tested by the author and Packt's editorial team. Packt does not accept AI-generated content that replaces expert authorship. Portfolio Director: Gebin George Relationship Lead: Sonia Chauhan Project Manager: Prajakta Naik Content Engineer: Afzal Shaikh Technical Editor: Rahul Limbachiya Indexer: Manju Arasan Production Designer: Shantanu Zagade Growth Lead: Nimisha Dua First published: May 2026 Production reference: 1270526 Published by Packt Publishing Ltd. Grosvenor House 11 St Paul's Square Birmingham B3 1RB, UK ISBN 978-1-80610-445-1 www.packtpub.com

📄 Page 4

Contributors About the authors Nestor Gandara is a technology leader with more than 20 years of experience in Generative AI, cloud, and digital transformation. As Principal Partner SA and Generative AI Strategist at Amazon Web Services, he works closely with C-suite executives to align emerging technologies with measurable business value. He also serves as a program lead, mentor, and learning facilitator for MIT Professional Education, and as a faculty member at IENYC and ISDI. In addition, he is the founder of NextLevel.guru and the author of The Art of Building Your Resilience and Adaptability. By combining executive leadership with talent mentoring, he operates at the intersection of next-generation enterprise technology, education, and business strategy. Eduardo Ordax is a Principal GenAI Go-to-Market at AWS with more than 15 years of experience across the technology industry, combining technical and business leadership with a strong focus on artificial intelligence. Recognized as the #1 AI influential voice in Spain and among the Top 20 worldwide, he is an international keynote speaker and postgraduate lecturer. He actively shares insights on AI innovation, strategy, and real-world adoption with a community of more than 200,000 professionals on LinkedIn, contributing to discussions shaping the future of artificial intelligence. Srikanth Daggumalli is a Senior Analytics & AI Specialist Solutions Architect at Amazon Web Services, specializing in generative AI, machine learning, and cloud-native data architectures. With nearly two decades of experience, he has designed mission-critical data platforms across industries including financial services, insurance, retail, automotive, ISVs, and digital-native enterprises. His expertise spans global payments, anti-money laundering, credit-risk management, and enterprise analytics for Fortune 500 organizations. He is an IEEE Senior Member, an IETE Fellow, and serves as a Technical Program Committee member and peer reviewer for IEEE conferences and Manning Publications. His technical articles published on InfoQ and the AWS Big Data Blog have been syndicated across more than 15 international platforms.

📄 Page 5

Ashutosh Dubey is a technology leader and recognized expert in Generative and Agentic AI at Amazon Web Services. He works closely with enterprise leaders to help operationalize artificial intelligence at global scale. A strong advocate for technical education and community engagement, he regularly shares industry insights through his public blogs and technical content. He is the coauthor of Generative AI for Software Developers and Interview Guide for Solution Architects, offering practical guidance for engineering and leadership teams navigating the complexities of modern software architecture and AI adoption.

📄 Page 6

About the reviewers Mona is a Senior AI/ML Specialist Solutions Architect at AWS with over 15 years of experience spanning AI/ML, cloud, and software engineering. She specializes in designing and scaling production-grade AI systems, including LLMs, fine-tuning, inference optimization, and enterprise AI architecture across AWS and Google Cloud. Mona is the author of two bestselling books and also a published speaker and researcher who has contributed to leading industry conferences and academic work. She actively mentors professionals in AI and cloud. She is an author of AI Agents on AWS. Sireesha Muppala, Ph.D. is a Senior Solutions Architecture Leader at Amazon Web Services with over 25 years of experience in digital transformation, cloud, and generative AI. She leads teams supporting Automotive and Manufacturing customers in adopting advanced cloud and AI technologies to drive business innovation. She has held leadership roles across organizations such as Oracle, Blackhawk Network, and Primer AI, and is an author of Amazon SageMaker Best Practices and Generative AI for Cloud Solutions. She is also an active speaker, researcher, and advocate for STEM education and underrepresented communities in technology. Satesh Sonti is a Principal Specialist Solutions Architect at AWS, specializing in enterprise data platforms, data warehousing, and AI/ML solutions. With over 20 years of experience, he has built and led complex data platform programs for banking and insurance clients globally, focusing on scalable and modern data architectures. He has developed a portfolio of more than 40 thought leadership assets, including blogs, reference architectures, workshops, and JAMs, and has delivered over 25 speaking engagements at major events such as AWS Summits and re:Invent. He is also a certified blog bar raiser and content guardian reviewer, contributing to the quality and consistency of technical content.

📄 Page 7

Subscribe for a free ebook New frameworks, evolving architectures, research drops, production breakdowns – AI_Distilled filters the noise into a weekly briefing for engineers and researchers working hands-on with LLMs and generative AI systems. Subscribe now and receive a free eBook, along with weekly insights that help you stay focused and informed. Subscribe at https://packt.link/8Oz6Y or scan the QR code below.

📄 Page 8

Table of Contents Preface xxv Free benefits with your book ............................................................................. xxx Chapter 1: The Generative AI Revolution – Why It Matters Now 1 The emergence of generative AI ............................................................................ 2 Generative AI trends • 4 The current state of generative AI ......................................................................... 6 Enterprise adoption trends • 6 The challenges of generative AI adoption • 8 Leveraging AWS tools for accelerated transformation • 9 Current use cases and business integration • 10 The technology behind the hype .......................................................................... 11 Advances in LLMs • 11 Infrastructure and data foundations • 16 The business imperative ...................................................................................... 17 Strategic advantages for early movers • 17 Risks of falling behind • 19 Navigating the hype vs. reality • 22 Responsible AI and trust • 23 Setting the stage for transformation ................................................................... 24 Summary ........................................................................................................... 25 References .......................................................................................................... 27 Chapter 2: How Generative AI Works 29 What sets generative AI apart from traditional AI? ............................................... 30 Machine learning vs. deep learning • 31 What are deep learning services on AWS? • 31

📄 Page 9

Neural networks explained simply ...................................................................... 32 What are neural networks used for? • 33 Foundation models and their training ................................................................. 36 What are foundation models? • 36 How they learn • 36 Foundation Model Evaluation and Iteration • 38 Standard benchmarks • 38 Why build your own evaluation • 39 Evaluation as an iterative process • 40 Navigating core architectures and adaptation techniques ..................................... 41 The building blocks • 41 Transformers • 41 Diffusion models • 42 Architecture • 43 Model adaptation strategies • 44 Pre-training: Build your model from scratch • 44 Approaches That Modify Model Weights • 45 Fine-tuning: Teach the model your language • 45 Approaches Without Changing Model Weights • 46 RAG (Retrieval-Augmented Generation): Bring in external knowledge • 46 Prompt engineering: Customize without code • 47 Guardrails, filters, and compliance • 48 Choosing the right adaptation strategy • 49 From cloud to capability ..................................................................................... 50 Amazon Bedrock and model access • 50 How to get started • 53 Practical examples • 54 Healthcare • 54 Marketing • 54 Education • 55 Legal workflows • 56 Small business automation • 57 Table of Contents viii

📄 Page 10

Empowering business and technical teams ......................................................... 58 Closing the language gap • 58 Picking the right tool for the right job • 59 Cross-functional enablement with AWS • 60 Executive leadership: Sponsorship to stewardship • 61 Summary ........................................................................................................... 62 References .......................................................................................................... 63 Chapter 3: The AWS Generative AI Stack – Tools of the Trade 65 The AWS AI stack ................................................................................................ 66 Infrastructure layer: The foundation for scalable GenAI ...................................... 67 Core AWS services in the infrastructure layer • 68 Amazon SageMaker AI • 69 Amazon Nova Forge: The foundation model factory • 69 AI compute: AWS Trainium and AWS Inferentia • 70 Data foundation for AI • 70 Amazon S3 (Data Lake and Vector Storage) • 70 Amazon SageMaker AI • 71 SageMaker feature store • 71 Data preparation and multimodal enrichment • 72 Amazon SageMaker HyperPod • 72 When to use infrastructure layer • 73 Foundation model and agent layer ...................................................................... 73 AWS services in the foundation model and agent layers • 74 Amazon Bedrock • 74 Model choice without Lock-in • 75 Benefits of model diversity • 75 Real-world use case • 75 Control and guardrails ........................................................................................ 76 AgentCore • 76 Nova Act • 78 Strands agents • 79 ix Table of Contents

📄 Page 11

Application layer ................................................................................................ 80 Core AWS services in the application layer • 81 Amazon Quick • 81 Quick is built around six core capabilities • 82 Spaces • 82 Chat agents • 83 Research • 83 QuickSight • 83 Flows • 84 Automate • 84 Built for enterprise security ................................................................................ 84 Quick in action • 85 Weekly business reviews • 85 Customer health monitoring • 85 Incident root cause analysis • 85 Onboarding playbook generation • 85 Amazon Quick Desktop: The future of work ........................................................ 85 Amazon Q Developer • 87 Kiro • 87 What makes Kiro stand out • 88 How Kiro works (in practical terms) • 88 Example scenarios • 88 AWS Transform • 89 Built for your biggest challenges • 90 AWS Transform Custom • 91 How AWS Transform works in practice • 91 Why this matters now • 92 From legacy to modern • 92 Amazon Connect • 92 AWS Marketplace • 94 Summary ........................................................................................................... 95 References .......................................................................................................... 96 Table of Contents x

📄 Page 12

Chapter 4: Data Readiness and Model Selections 97 Preparing data aka "fuel" ................................................................................... 98 Data challenges and architectures • 99 AWS analytics services • 102 Amazon SageMaker AI: AWS's purpose-built platform for AI and ML .................. 110 Magic of the Apache Iceberg REST API layer • 115 AWS generative AI stack • 117 Building with Amazon Bedrock .......................................................................... 118 Choosing models as per your needs • 119 Decide model and evaluation processes .............................................................. 121 LLM as a judge • 122 Amazon Bedrock Evaluations • 122 Python LangChain Model Laboratory • 123 Other evaluation metrics • 124 Script and automate the process • 126 Summary .......................................................................................................... 128 References ......................................................................................................... 129 Chapter 5: Build an Agentic GenAI Application 131 Agentic GenAI application architecture .............................................................. 132 Technology Stack .............................................................................................. 134 AI and agent frameworks • 134 Amazon Bedrock AgentCore • 134 Strands Agentic SDK • 134 Amazon Bedrock with Claude Opus 4.7 • 134 Database and Storage • 134 PostgreSQL 15 • 134 Web interface and front end • 135 Prerequisites • 135 AWS account and access • 135 Python libraries • 135 xi Table of Contents

📄 Page 13

A note on Model Context Protocol (MCP) • 136 Architecture flow .............................................................................................. 137 Hotel discovery tools • 138 Implementation guide ....................................................................................... 139 Prerequisite setup • 139 Step 1: Project structure setup • 140 Step 2: Install dependencies • 142 Step 3: Environment configuration • 143 Step 4: Database initialization • 143 Step 5: Agent configuration • 151 Step 6: Build frontend user interface with Streamlit application • 152 Step 7: Build agent tools • 154 Summary .......................................................................................................... 183 References ......................................................................................................... 183 Chapter 6: Production-Ready Agentic & GenAI Applications 185 Deployment strategies ...................................................................................... 186 Understanding deployment options: Containerization with Docker • 186 Creating Docker images for your GenAI application • 186 Multi-stage builds for optimization • 186 Managing dependencies and environment variables • 187 Application deployment options ........................................................................ 187 Amazon ECS (Elastic Container Service) for container orchestration • 188 AWS Fargate for serverless container deployments • 188 Amazon EKS (Elastic Kubernetes Service) for complex workloads • 188 AWS Lambda for event-driven architectures • 189 AWS Lambda Durable Functions • 190 Model inference deployment options ................................................................. 192 Amazon Bedrock • 192 Foundation Models • 192 Custom Model Import • 193 Amazon SageMaker AI endpoints for custom models • 193 Table of Contents xii

📄 Page 14

When SageMaker makes sense • 193 Self-Hosted models • 194 When Self-Hosting is preferred • 194 Deploying Infrastructure as Code (IaC) with AWS CDK or CloudFormation ......... 195 Choosing between CDK and CloudFormation • 196 CI/CD pipelines • 197 Scaling GenAI applications ............................................................................... 198 Horizontal scaling • 199 Auto scaling groups configuration • 199 Load balancing with Application Load Balancer (ALB) • 200 Handling concurrent requests efficiently • 201 Vertical scaling considerations • 201 Right-sizing compute resources (CPU, memory, GPUs) • 201 Amazon Bedrock throughput and provisioned capacity • 202 SageMaker endpoint scaling strategies • 202 Performance optimization • 203 Response streaming for better user experience • 203 Batch processing for high-volume workloads • 203 Connection pooling for database optimization • 204 Cost optimization • 204 Monitoring and managing inference costs • 204 Using Spot Instances where appropriate • 204 Implementing request throttling and rate limiting • 205 Model selection based on cost-performance tradeoffs • 205 Caching strategies (Amazon ElastiCache, Redis, CloudFront) • 206 Monitoring and observability • 206 Application monitoring • 207 Custom metrics for GenAI-specific KPIs (latency, token usage, error rates) • 209 AWS X-Ray for distributed tracing • 210 Model Performance Tracking • 210 Alerting and incident response • 213 Security best practices ....................................................................................... 214 xiii Table of Contents

📄 Page 15

Authentication and authorization • 215 Amazon Cognito for user authentication • 215 IAM roles and policies for service-to-service communication • 216 API Gateway with authentication mechanisms • 217 Fine-grained access control • 217 Data security • 218 Encryption at rest • 218 Encryption in transit (TLS/SSL) • 218 VPC configuration and network isolation • 218 Secrets management with AWS Secrets Manager • 219 GenAI-specific security • 219 Prompt injection prevention techniques • 219 Input validation and sanitization • 220 Output filtering and content moderation • 221 Amazon Bedrock Guardrails implementation • 222 Amazon Bedrock AgentCore for production-ready agentic AI ............................. 224 The runtime component • 226 A serverless execution environment for any framework • 226 Model flexibility • 227 Multimodal and multi-agent support • 227 Built-in identity integration • 228 Session isolation • 228 The gateway component • 228 Making services accessible to agents • 228 Prebuilt integrations • 229 MCP server connections • 229 The memory component • 229 Context that persists across sessions • 229 Short-term and long-term memory • 230 Shared memory and learning • 230 The identity component • 231 Enterprise-grade authentication for agents • 231 Table of Contents xiv

📄 Page 16

Session security and management • 231 The policy component • 232 Deterministic control over agent behavior • 232 Business rule enforcement • 232 The observability component – unified visibility • 232 The evaluation component • 233 The browser component • 234 Managed web automation • 234 Real-world use cases • 234 The code interpreter component • 234 Safe code execution in isolated environments • 235 Integration with your Strands application • 235 The registry component • 235 The payments component • 236 Summary ......................................................................................................... 236 References ......................................................................................................... 237 Chapter 7: Advanced Techniques and Emerging Trends 239 Agent-to-Agent communication: Beyond centralized orchestration ................... 240 When supervisors become bottlenecks • 240 Agent Mesh Architecture with Amazon Bedrock AgentCore • 241 The foundation: Managed agent execution • 242 Agent discovery with A2A protocol • 242 Multi-agent collaboration via AgentCore Runtime • 243 Extending with Event-Driven patterns (custom architectures) • 244 Complete flow example with AgentCore • 245 Infrastructure cost and scale • 246 Advanced pattern: Agent negotiation • 247 Framework flexibility with AgentCore • 248 Advanced reasoning: Tree of thoughts and Reflexion ......................................... 248 Tree of thoughts: Systematic multi-path exploration • 249 The Tree of Thoughts solution • 250 xv Table of Contents

📄 Page 17

The initial deliberation • 250 Pruning deep dive • 251 The strategic synthesis • 251 Cost vs. accuracy trade-off • 253 Reflexion: Agents that learn from failures • 254 The Reflexion pattern • 255 Implementation with AgentCore and OpenSearch • 256 Hybrid Neuro-Symbolic systems • 258 Knowledge graphs plus LLMs • 258 The hybrid solution • 258 Query flow example • 258 Cost and accuracy comparison • 260 Rule engines plus generative AI • 262 Multi-stage processing solution • 263 Specialized agents: Computer use and voice ...................................................... 265 Computer use with Amazon Nova Act • 265 Implementation with Amazon Bedrock AgentCore and the browser tool • 268 Security framework • 270 Cost analysis and ROI • 270 Decision flow for computer use • 270 Voice agents with Amazon Nova Sonic • 272 The foundation: STT and TTS models • 272 The cascading latency problem • 272 The modern approach: Bidirectional streaming • 273 Implementation with the Strands framework • 273 Architectural shift: From sequential to streaming ............................................... 274 Conversation example • 274 Latency optimization strategies • 276 Cost analysis and ROI • 277 Strategic decision matrix for voice AI ................................................................ 278 Production patterns at scale • 279 Hierarchical memory management • 279 Table of Contents xvi

📄 Page 18

Three-tier memory solution • 280 How it works in practice • 280 Implementation with Amazon Bedrock AgentCore • 281 Distributed tracing with AWS X-Ray • 281 Example trace • 281 Constitutional AI for production safety • 282 Constitutional AI patterns • 282 Example execution • 283 Emerging trends: 12–24 month horizon ............................................................. 284 Test-time compute and intelligent scaling • 285 Cross-framework interoperability via AgentCore • 285 Autonomous agent evolution • 286 Summary ......................................................................................................... 286 Chapter 8: From Strategy to Use Case – Identifying Business Value 289 Matching GenAI to business needs .................................................................... 290 Strategic goals, not just cool demos • 291 Know when to use and not use GenAI • 293 Traditional solution was sufficient • 293 Generative AI was a game changer • 293 AI as a partner, not just a tool • 294 Rethinking Human-AI Collaboration • 295 Use case: A global consulting firm • 295 From idea to impact • 297 Exploring GenAI potential across industries • 298 Retail – personalized in-store recommendations • 298 Healthcare – multilingual patient education • 299 Financial services – Compliance-ready meeting notes • 299 Hospitality – personalized guest experiences at scale • 300 Manufacturing – equipment maintenance summary • 300 Designing a GenAI platform for multiple use cases • 301 Future-proofing your AI initiatives • 301 xvii Table of Contents

📄 Page 19

Automated Reasoning Checks (Amazon Bedrock Guardrail) • 302 Utility Incident Management • 303 A multi-domain GenAI platform • 304 The RAG strategy: managed vs. custom • 308 The managed path - Amazon Bedrock Knowledge Base • 308 Measuring success and continuous improvement • 310 AWS-native capabilities for tracking and optimization • 310 Common pitfalls and how to avoid them • 312 Summary .......................................................................................................... 315 Chapter 9: Industry Spotlights and Case Studies 317 Generative AI in business intelligence and analytics ........................................... 318 Natural language interfaces for data • 319 How does an NLI work? • 319 NLIs in practice • 320 From queries to insights with Amazon Q in QuickSight • 321 Case study: Generative BI for a retail giant • 321 Business problem • 322 The architectural solution • 322 Generative AI in creative industries ................................................................... 325 The challenges that GenAI addresses • 325 Case study: Sports entertainment • 326 Business problem • 326 The architectural solution • 326 Measurable strategic payoffs • 328 Four architectural lessons for practitioners • 329 Generative AI in marketing ............................................................................... 330 Accelerating creative output and asset production • 330 Dynamic campaign analytics and performance optimization • 330 Case study: Marketing design • 331 Business problem • 331 The architectural solution • 332 Table of Contents xviii

📄 Page 20

Measurable strategic payoff • 334 Generative AI in healthcare and life sciences ..................................................... 334 Streamlining administrative workflows • 335 Case study: Clinical trial documentation • 335 Business problem • 336 The architectural solution • 336 Measurable strategic payoffs • 338 Broader applications across the industry • 339 Free-form text analysis • 339 Clinical trend identification • 339 Accelerated medical image analysis • 339 Accelerated research • 340 Generative AI in customer service and operations .............................................. 340 Integrating Amazon Connect and Amazon Q • 341 Case study: Augmented customer support • 342 Business problem • 342 The architectural solution • 343 Measurable strategic payoff • 344 Case study: Autonomous platform engineering with Bedrock AgentCore ........... 345 Business problem • 345 The architectural solution with Amazon Bedrock AgentCore • 345 Measurable Strategic Payoffs • 347 Upcoming trends in industry-specific generative AI ........................................... 348 Healthcare: Precision medicine and proactive care • 348 Finance: Hyper-personalized financial advisory and risk management • 348 Manufacturing: Intelligent mass production to mass customization • 349 Retail and e-commerce: Hyper-individualized shopping experiences • 349 Education: Adaptive learning and personalized tutoring • 349 Entertainment: Content curated and created for you • 350 The path forward: Innovation and governance • 350 Summary ......................................................................................................... 350 xix Table of Contents