GenAI on Google Cloud. Enterprise Generative AI Systems and Agents Ayo Adedeji, Lavi Nigam, Sarita Joshi

Dostępność produktu

Produkt nie został jeszcze oceniony pod kątem ułatwień dostępu lub nie podano żadnych informacji o ułatwieniach dostępu lub są one niewystarczające. Prawdopodobnie Wydawca/Dostawca jeszcze nie umożliwił dokonania walidacji produktu lub nie przekazał odpowiednich informacji na temat jego dostępności.

Spis treści książki

• Preface
  • Why This Book Matters
  • What Youll Find in This Book
  • Our Approach
  • Who This Book Is For
  • Prerequisites
  • Conventions Used in This Book
  • Using Code Examples
  • OReilly Online Learning
  • How to Contact Us
  • Acknowledgments
• 1. The Challenge of Generative AI Application Development
  • Overview of LLMs, Generative AI Agents, and Potential Applications to Business Tasks
    • Small Language Models (SLMs)
    • Foundation Models and Multimodality
    • Domain-Specific and Reasoning Models
    • Generative AI Agents
    • Agent Architectures
  • Challenges in Development, Deployment, and Maintenance
    • Development Challenges
      • Data
      • Models
      • Evaluation
    • Deployment Challenges
      • Infrastructure
      • Integration
      • Security
    • Maintenance Challenges
      • Drift
      • Monitoring
      • Version control
    • Addressing Challenges with Modern Platforms
  • Industry Use Cases and ROI
  • Looking Ahead
  • Learning Labs
• 2. Data Readiness and Accessibility
  • The Amplified Importance of Data for GenAI
  • What Data Readiness Really Means for GenAI Applications
    • Key Dimensions of Data Readiness
    • The Interconnected Nature of Data Readiness
    • Managing Prompts as Data Assets
  • The Human Element: Roles in the Data Readiness Journey
    • Data Scientists: The Explorers
    • ML Engineers: Building the Bridge to Production
    • Data Engineers: Architecting the Foundation
    • DevOps and SREs: Operationalizing the Foundation
    • Business SMEs and Domain Leaders: The Why Behind the What
  • Strategic Data Patterns: The Foundation for Reliable GenAI Systems
    • The Unified Data and AI Platform
    • From RAG to Agentic RAG: The Evolution of a Data Pattern
      • The DIY RAG pipeline: a data engineering challenge
      • The RAG trade-off: choosing your vector store (DIY versus managed)
      • Vector store: choosing the right approach
      • Advanced RAG: from vectors to agents
      • GraphRAG: retrieving relationships, not just documents
      • Agentic RAG: orchestrating specialist retrievers
    • Tying it All Together: the Enterprise RAG Knowledge Engine
      • Phase 1: Building the Knowledge Engine (the Data Foundation)
      • Phase 2: Activating the Agentic RAG layer
  • Data Readiness for Agent Systems
  • Security and Governance: Protecting Data Throughout the LLM Lifecycle
    • Data Privacy Framework
    • Comprehensive Governance
  • Practical Data Readiness Assessment
  • Looking Ahead
  • Learning Labs
• 3. Building a Multimodal Agent with the Agent Development Kit (ADK)
  • From Zero to Agent in Seven Lines
    • The Simplest Thing That Works
    • The Runtime Behind the Simplicity
    • Running Your First Conversation
    • Understanding the Limitations
  • Adding Intelligence Through Tools
    • Your Agents First Tool
    • Tools Versus SubagentsA Practical Decision Framework
  • State Management That Actually Scales
    • Building a Stateful Shopping Cart
    • Understanding the Three Scopes
    • State Scope Interactions
    • Making State Persist in Production
      • Development: InMemorySessionService
      • Self-managed production: DatabaseSessionService
      • Fully managed production: VertexAiSessionService
  • Beyond Structured State: Semantic Memory
    • Vertex AI Agent Engine Memory Bank: Learning from Conversations
    • Implementation
      • Step 1: Add memory service to your runner
      • Step 2: Extract memories from completed sessions
      • Step 3: Retrieve memories automatically
      • Structured state and semantic memory working together
  • Expanding to Multimodal
    • Making Our Agent See
    • From Static Analysis to Live Support
    • Building Complete Interaction Memory
    • Building Production-Grade Tools
    • Handling Asynchronous Operations
    • Ensuring Safety with Human-in-the-Loop
    • Production Monitoring and Policy Enforcement with Callbacks and Plug-ins
  • Looking Ahead
  • Learning Labs
• 4. Orchestrating Intelligent Agent Teams
  • The Bottleneck of the Monolithic Agent
    • Conflicting Instructions
    • Tool Selection Paralysis
    • Token Limitations
    • Maintenance Nightmare
    • The Solution: An Agent Team
  • The Roadmap: From Local Teams to Distributed Systems
  • Local Teams
    • The Foundation: Agent Hierarchy
    • Pattern 1: The Assembly Line (SequentialAgent)
    • Pattern 2: The Independent Taskforce (ParallelAgent)
    • Pattern 3: The Iterative Refiner (LoopAgent)
    • Distributed Collaboration
    • The Organizational Why
    • MCP: The Language of Tools
    • A2A: The Language of Delegation
      • Exposing a specialist agent: the finance teams perspective
      • Consuming the specialist agent: the customer support teams perspective
    • Putting It All Together: A Hybrid Agent Team
    • Production Realities
    • The Trust Problem: Security Schemes in A2A
    • The Extension Problem: Evolving Agent Capabilities
    • The Visibility Problem: Distributed Tracing
    • The Versioning Problem: Managing Agent Evolution
  • Looking Ahead
    • Edge and Embodied Intelligence
    • From Architecture to Excellence
  • Learning Labs
• 5. Evaluation and Optimization Strategies
  • Tailoring Evaluation to Your LLM/Agents Purpose
    • Beyond Basic Functionality
    • Key Dimensions of Evaluation
      • Quality of output
      • Task success and goal achievement (agents)
      • System performance
      • Robustness and reliability
      • Safety and responsibility
    • Setting the Bar for Production Excellence
  • Practical Evaluation Strategies
    • Human-Centered Evaluation
    • A/B Testing and Preference Scoring
      • Side-by-side comparison
    • Red Teaming: Stress Testing for Safety and Reliability
    • Automated Evaluation: Scaling Feedback for Rapid Improvement
    • Reference-Based Metrics for Text Generation
    • Limitations of Reference-Based Evaluation
    • Domain-Specific and Task-Oriented Metrics
    • Metrics for Agentic Systems and Tool Use
      • Understanding the agent evaluation process
      • Trajectory evaluation and metric visualization
      • Custom metric development for advanced evaluation
        • LLM-based custom metrics
        • Computation-based custom metrics
  • Optimization Strategies
    • Refining Prompts
    • Elevating Agent Performance
      • Designing robust tools
      • Improving agent reasoning and planning
      • Multiagent optimization patterns
    • Beyond Prompt and Agent Optimizations
  • Looking Ahead
  • Learning Labs
• 6. Tuning and Infrastructure
  • The Tuning Decision
  • The Fine-Tuning Decision Framework
    • Fine-Tuning Strategies: From Full Training to Efficient Adaptations
    • The Real Cost of Fine-Tuning
    • Implementation Approaches
    • Infrastructure Questions Emerge
    • The Constraint Youll Hit First
    • Pattern 1: The Waiting Accelerator
    • Pattern 2: The Memory Wall
    • Pattern 3: Maxed Out But Still Slow
    • Pattern 4: More GPUs = Worse Performance
    • Accelerators: Matching Hardware to Bottlenecks
    • The Decision Framework
    • The Practical Decision
    • Migration Reality
    • Storage Options
    • When Storage Becomes Your Bottleneck
    • The Storage Pattern
      • Training phase: Managed Lustre
      • Archival phase: Cloud Storage
      • Serving phase: Cloud Storage FUSE + Anywhere Cache
    • Serving and Deployment
      • Efficient serving with vLLM
        • The throughput problem
        • PagedAttention
        • Continuous batching
        • Disaggregated serving
    • Configuration That Matters
      • Tuning to your bottleneck
      • Multiplatform support
    • Connecting Models to Agents
    • Agent Deployment Platforms
    • Agent Engine
    • Cloud Run
    • GKE
      • Google Distributed Cloud
      • Platform comparison
  • Looking Ahead
  • Learning Labs
• 7. MLOps for Production-Ready AI and Agentic Systems
  • From Ad Hoc to Systematic: The Current State of Teams
  • The Evolution of MLOps
  • Building Reproducible Training Pipelines
    • Data Versioning and Lineage
      • Training data
      • Context data
      • Prompt templates
      • Data lineage
    • Experiment Tracking
    • Model Registry and Governance
    • Automated Retraining
  • Comprehensive Monitoring
    • Agent Monitoring
      • Cloud Trace for agent execution
      • Token usage and cost attribution
      • Loop and iteration visibility
      • Custom monitoring with Prometheus
    • Technical Monitoring
    • Hallucination Detection
  • CI/CD for AI Systems
    • Cloud Build
    • Cloud Deploy
  • Security and Governance as Foundation
    • Security Framework for AI Agents
    • Model Armor: A Key Security Component
  • Cost Management
    • The True Cost Model
    • Cost Attribution Strategies
    • Intelligent Cost Operations
      • FinOps hub with Gemini Cloud Assist
      • Cloud Billing reports with AI assistance
      • FinOps hub for optimization intelligence
      • Beyond console insights
    • Spending Controls
      • User-level controls
      • Technical circuit breakers
  • Looking Ahead
  • Learning Labs
• 8. The AI and Agentic Maturity Framework
  • What Is the AI and Agentic Maturity Framework?
  • The Maturity Dimensions and Phases
    • Vision and Leadership (The What and the Why Dimension)
      • Tactical phase
      • Strategic phase
      • Transformational phase
      • Example: breaking out of the Tactical phase
      • Advice from the authors
    • Talent and Culture (The Who Dimension)
      • Tactical phase
      • Strategic phase
      • Transformational phase
      • Example: moving beyond the Strategic phase
      • Advice from the authors
    • Operational and Technical Practice (The How Dimension)
      • Tactical phase
      • Strategic phase
      • Transformational phase
      • Example: an organization operating in the Transformational phase
      • Advice from the authors
    • How the Three Dimensions of AI and Agentic Maturity Can Work Together
  • From Framework to Reality: What Are Teams Actually Building, and How?
    • Technical Conversations
    • Leadership, Talent, and Culture Conversations
    • Why and How a Platform Approach Can Accelerate an Organizations AI and Agentic Maturity
    • Vertex AI Platform
      • Enterprise governance: from isolated code to auditable lineage
      • Seamless integration: embedding AI as a business asset
      • Production velocity: from manual hand-off to AIOps engine
      • Gemini Enterprise
  • Learning Labs
• Conclusion
• A. Further Reading for Leaders
• Index

pokaż cały spis treści

ukryj recenzje