RAG-Ready Patterns for Data Platforms. Making Systems Ready for Grounded Generative AI Ravi Vedula, Gerardo Bodegas Martinez, Maruti Chittajallu

Dostępność produktu

Produkt został zweryfikowany pod kątem ułatwień dostępu więcej przeczytasz w deklaracji dostępności.

Spis treści książki

• Preface
  • Why We Wrote This Book
  • Who Should Read This Book
  • What This Book Is and Is Not
  • Navigating This Book
  • Conventions Used in This Book
  • Using Code Examples
  • OReilly Online Learning
  • How to Contact Us
  • Acknowledgments
• 1. The Pillars of RAG Readiness
  • Why the Gap Exists
  • Why RAG-Ready Patterns Are Needed Now
  • A Practical RAG-Readiness Framework
    • Pillar 1: Data Assets
      • Case study: Three definitions of a user
      • The Entity Kernel Pattern
    • Pillar 2: Infrastructure
      • Case study: Fixing yesterdays numbers
    • Pillar 3: Trust Layer
      • Case study: Trust at scale in IDEAS
    • Quick Self-Assessment Across All Pillars
  • Conclusion
• 2. The Semantic Backbone: Designing Data Assets for RAG
  • Identify the Entity Kernel
  • Capture Source Semantics in Data Contracts
  • Govern Semantic Change
  • Organize Assets into Layered Semantic Asset Taxonomy
    • Source-Boundary Assets
    • Entity-Kernel Assets
    • Activity-Grain Assets
    • Dimensions
    • Metrics
    • Segments
    • Derived Serving Artifacts
  • Use Portable, Declarative Definitions
  • Make Metadata Available Through Suitable APIs
  • Putting the Patterns Together
  • Conclusion
• 3. Metadata-First Pipeline Engineering
  • Semantic Definitions Arent Enough
    • The Enforcement Gap
    • When Data Lies
  • The Metadata-First Mindset
  • The Pipeline Contract
    • Schema and Semantic Definitions
    • Operational Guarantees
    • Governance
    • Lineage
  • Pipelines-As-Code Model
    • Why Code Wins
    • Developer Experience
    • Pattern Example
  • Managed Fabric Code Pattern
    • Managed Data Fabric Responsibilities
    • Building a Pipeline Example
      • 1. Connecting to data with a connector
      • 2. Transforming data with an activity
      • 3. Sharing data safely with a view
      • 4. Testing your code with a unit test
      • 5. Registering the pipeline
  • Live Metadata for RAG
    • Smarter Retrieval
    • Self-Healing Context
  • Conclusion
• 4. Data Quality As a Retrieval Signal
  • Beyond the Enforcement Gap
    • When Plausible Data Misleads RAG
    • Why Intent Is Not Enough
  • Activating the Contract
    • Defining the Quality Signal
    • Retrieval Policies
  • The Quality Engine
    • Generating the Signal
    • Monitoring As Code
    • Continuous Validation and Gatekeeping
    • The Quality Signal As a Product
  • The Quality Signal Protocol
    • Standardizing Trust
    • The Schema of Trust
    • Exposing State, Not Explanations
    • Design Constraints for Enterprise Scale
    • Pattern Example
  • Signal-Aware Retrieval
    • Compositional Trust
      • Unified trust policies
      • Partial context and omission
    • Signal-Driven Prompt Engineering
      • From retrieval policy to prompt policy
      • Using diagnostic context
    • The Trust-Relevance Trade-Off
    • Trust-Aware Orchestration in Practice
  • Establishing a Culture of Accountability
    • From Implicit to Visible Trust
    • From Ownership to Action
    • The Virtuous Cycle of Trust-Aware Systems
  • Conclusion
• 5. Curated Grounding: Golden Answers and the Living Glossary
  • Why Clean Data Alone Doesnt Answer Business Questions
  • Challenges Specific to LLMs on Enterprise Data
    • Data Asset Semantics
    • Data Asset Ownership and Management
      • Lack of asset contracts and versioning
      • Mismatched scenario descriptions
      • Undocumented domain knowledge
      • Overlapping assets
      • Inadequate traceability erodes trust
    • A Strategy Built on Curated Grounding
  • Golden Answers
    • The Golden Answer Validation Workflow
      • Sourcing questions
      • Roles and governance
      • The rest of the workflow at a glance
    • Schema, Metadata, and Versioning
    • Evaluation and Monitoring
  • The Living Glossary
    • Why a Glossary Matters for Retrieval
    • Governance and Maintenance
      • Clear ownership
      • Lifecycle and workflow
      • Start small, then expand
    • Schema
  • Accelerating Curation with AI: Scaling Human-in-the-Loop Review
    • Candidate Authoring
    • Using LLMs As Graders
    • Linking, Routing, and Cross-Step Maintenance
    • Production Monitoring and Backlog Feedback
  • Conclusion
• 6. Scenario-Based Access Control for RAG Systems
  • The Access Control Paradox
    • Context and Least Privilege
    • When Citation Becomes Exposure
    • The Governance Enforcement Gap
  • From Who to Why: The Scenario-Based Access Pattern
    • Defining the Access Scenario
    • Approved Scenarios
      • Preapproved baseline scenarios
      • Restricted scenario enrollment
    • Scenario-Based Retrieval Scope
    • Scenarios As the Governance Filter
    • Scenario Eligibility
  • The Access Scenario Specification
    • The Governance Handshake
    • Core Fields
      • business_justification
      • owner
      • compliance_approvers
      • personas
      • eligibility
      • asset_neighborhood
      • scenario_visibility
      • lifecycle
    • The Specification in Practice
    • Scenarios and Privacy Tags
  • Enforcing Scenarios: The Secure Retrieval Filter Pattern
    • The Governance Filter
    • Compositional Governance and the Intersection Rule
    • Dynamic Redaction and Policy Enforcement
    • Pattern Example: The Secure Retrieval Filter
  • Governing AI Responses
    • Intent-Aware Relevance
    • Graceful Denial and User Guidance
    • Auditability and the Reason for the Response
  • Conclusion
• 7. Building RAG-Ready Data Products
  • Beyond the Gold Layer and the Data Mesh
    • Where Data Mesh Raises a Different Problem
      • The domain boundary problem
      • The coordination tax
      • The consumer perspective
      • The N-incompatible-interfaces problem
    • Isnt This Just Data Mesh?
  • What a RAG-Ready Data Product Actually Is
    • The Product Boundary Question
    • The Running Example: Commercial Product Usage
  • Serving Every Modality: From Natural Language to Code
    • Multidepth: Same Product, Different Personas
    • Structured: The Primary Modality
      • Query envelope definition
      • Grounding example integration
      • The business-logic-in-presentation problem
    • Metadata: The Definitional Modality
    • Unstructured and Computed Modalities
  • Versioned Contracts for Trust: The Federated Data Product Facade
    • The Federated Data Product Facade Architecture
      • The grounding portal is adjacent, not composed
      • The facade reflects the platform enforces, domain contributes principle
    • The Facade Specification
    • The Facade As Orchestration Backend
    • Versioned Structure, Live Signals
    • Building the Facade Incrementally
  • Supplying Business Meaning to the Facade: The SME as Coder
    • Declarative Definitions As Product Metadata
    • The SME Authoring Experience
    • The Grounding Portal As the Bridge
    • Declarative Definitions As Retrieval Targets
  • How RAG-Ready Data Products Are Discovered, Selected, and Maintained
    • The Discovery Problem Revisited
    • Topic Taxonomy: Organizing Discovery Around Business Questions
    • How Selection Actually Works
    • The Data Product Manager As the Human Facade
    • The Data Product Lifecycle
  • Conclusion
• 8. Orchestrating RAG with the Model Context Protocol
  • The Universal Connector: An Introduction to MCP
    • Why MCP Is Gaining Adoption
      • Utilizes your existing capabilities
      • Provides rapid AI enablement
      • Retains HITL review
      • Supports focused agents and workflows
      • Reuses MCP tools across AI vendors
      • Allows global implementations
    • Pitfalls to Be Aware Of
  • Exposing Data Products As MCP Services
    • Control and Trust
    • Discovery, Provenance, and Semantics
    • Data Operations
    • Platform Operations
  • Building Transparent RAG Chains with Human-in-the-Loop Participation
    • Example 1
    • Example 2
    • Example 3
    • Example 4
    • Example 5
    • Example 6
      • Example Takeaways
  • Guidelines and Recommendations for MCP RAG Chains
    • Design MCP Tools with a Security-First Posture
    • Observability As a Core Design Goal
    • Impose Discipline to Prevent Tool Sprawl
    • Govern Workflows, Not Just Tools
    • Isolate MCP Integrations from Core Business Logic
  • A Playbook for Agentification with MCP
    • Baseline
      • Stage 1: Establish a common MCP sequence repository
      • Stage 2: Create parameterized instruction templates
      • Stage 3: From templates to skills
    • Where to Start and When to Advance
  • From Observability to the Platforms Nervous System
    • Observability from MCP Tools
    • Key Metrics Sourced from MCP Usage
  • From Signals to Improvements
  • Conclusion
• 9. Designing the Enterprise RAG Stack
  • From Prototype to Production: The Enterprise Imperative
    • The Ten Challenges
      • Query Understanding: Challenges of understanding the question
      • Adaptive Retrieval: Challenges of finding the answer
      • Response Assurance: Challenges of trusting the answer
    • Lessons Learned
  • The Enterprise RAG Lifecycle: A Reference Architecture
    • How the Lifecycle Builds on Earlier Chapters
    • Where MCP Tools Slot In
    • The IDEAS Copilot v2 as an Instance of the Lifecycle
    • The Enterprise RAG Capability Manifest
    • The Full Stack: Control Plane, Serving Plane, and Feedback Loop
  • The Query Understanding Capability
    • Intent Classification: From Bespoke Routing to Grounding-Driven Disambiguation
    • The Question-to-Asset Ontology
    • Entity Resolution on a Data Platform
    • Temporal Grounding
  • The Adaptive Retrieval Capability
    • Multimodal Retrieval over a Data Platform
    • Grounding-Driven Query Generation
    • Quality-Gated Retrieval
  • The Response Assurance Capability
    • Citation-First Response Generation
    • Guardrails and Graceful Denial
    • The Feedback-Driven Improvement Loop
  • Designing for Modularity and Safety
    • Lineage-Aware Retrieval and Data Versioning
    • Schema-Driven Grounding and Query Generation
    • Metric Certification Trust Chains
    • Governance-Integrated Synthesis
    • Latency and Cost Budgets
    • Evals As an Operating Discipline
      • Evals traverse the RAG stack
      • From grounding to the eval suite
  • The Living Architecture
  • Conclusion
• 10. RAG-Readiness Scorecards
  • What the Scorecards Diagnose
  • A Common Assessment Discipline
  • Facade Completeness Index
    • How the FCI Works
    • What the FCI Result Reveals
    • Reading the Portfolio, Not Just the Percentage
  • Definition Governance Index
    • How the DGI Works
    • Choosing and Maintaining the Definition Inventory
  • Grounding Coverage Matrix
    • How the GCM Works
    • What the GCM Result Reveals
    • Coverage, Depth, and Validation Are Separate Questions
  • Staleness Detection Framework
    • How the SDF Works
    • Applying the Lowest-Wins Rule in Practice
    • Consequences by Artifact Class
    • What the SDF Result Reveals
  • RAG Failure Diagnostic Tree
    • How the RFDT Works
    • Branch 1: Wrong Answer
    • Branch 2: No Answer
    • Branch 3: Slow Answer
    • Branch 4: Inconsistent Answer
    • Branch 5: Unauthorized or Noncompliant Information
    • What a Validation Record Must Contain
    • Coverage Does Not Equal Effectiveness
  • Feedback Conversion Rate
    • How the FCR Works
    • What the FCR Result Reveals
  • The Composite RAG-Readiness Profile
    • Establishing the Assessment Frame
    • Normalizing Without Erasing Meaning
    • Reading the Shape
    • Compact AnalyticsCo Illustration
    • The Profile As Shared Diagnostic Language
      • What the profile can say
      • What the Profile Cannot Say
  • From Profile to Plan
    • Reading the Profile As a Set of Findings
    • What the Profile Orders and What It Does Not
    • Routing Findings to Established Patterns
    • When the Profile Expires
  • Conclusion
• 11. Practices and Pitfalls: Building the Future of RAG-Ready Enterprises
  • Practices That Always Pay Off
    • Practice 1: Start Narrow, Expand Through Metadata
    • Practice 2: Make Metadata the Medium of Collaboration
    • Practice 3: Emit Machine-Readable Trust Signals
    • Practice 4: Govern Purpose, Not PeopleScenario-Based Access Control
    • Practice 5: Run Governance as Infrastructure
    • Practice 6: Let Feedback Loops Refactor the Platform
    • Practice 7: Distribute Ownership, Centralize Standards
  • Pitfalls That Derail Trust
    • Pitfall 1: The Ontology Trap
    • Pitfall 2: Code-First Chaos
    • Pitfall 3: Shadow Governance
    • Pitfall 4: Hallucination by Design
    • Pitfall 5: The Hero Team Syndrome
    • Pitfall 6: Neglecting Human-in-the-Loop Reviews
    • Pitfall 7: Measuring Vanity, Not Value
  • Proving Value in Increments
    • Anchor on Outcomes, Not Platforms
    • Measure What Builds Trust
    • Turn Success into Habit
    • Make Progress Legible to Leadership
    • Avoid the Trap of Over-Promising
  • Looking Ahead
    • From Analytics Platforms to Conversational Substrates
    • The Rise of Agentic Retrieval
    • Federated Trust, Runtime Governance
    • Sustainable Intelligence at Scale
  • Conclusion: From Readiness Gap to Readiness Mindset
• Index

pokaż cały spis treści

ukryj recenzje