Name: Context Engineering for Multi-Agent Systems
Author: Denis Rothman
ISBN: 9781806690046

微信读书书城Context Engineering for Multi-Agent Systems

首页我的书架登录

翻译完成

切换

Context Engineering for Multi-Agent Systems

Denis Rothman

扉页
版权信息
+书签
Dedication
Contributors
Preface
1 From Prompts to Context: Building the Semantic Blueprint
Understanding context engineering
Level 1: The basic prompt (zero context)
Level 2: The better context (linear context)
Level 3: The good context (goal-oriented context)
Level 4: The advanced context (role-based context)
Level 5: The semantic blueprint
SRL: from linear sequences to semantic structures
Building an SRL notebook in Python
The main function: visualize_srl
Defining the semantic roles
The plotting engine: _plot_stemma and canvas setup
Dynamic positioning and drawing the stemma (graph)
Running SRL examples
Engineering a meeting analysis use case
Layer 1: Establishing the scope (the “what”)
Layer 2: Conducting the investigation (the “how”)
Layer 3: Determining the action (the “what next”)
Summary
Questions
References
Further reading
2 Building a Multi-Agent System with MCP
Architecting the MAS workflow with MCP
Building an MAS with MCP
Initializing the client
Defining the protocol
Building the agents
Building the Orchestrator
Running the system
Error handling and validation
Building robust components for the LLM
Validating MCP messages
Adding agent specialization controls and validation
The final Orchestrator with a validation loop
Running the final robust system
The evolution of AI architecture
Tools for building agent systems
Summary
Questions
References
Further reading
3 Building the Context-Aware Multi-Agent System
Architecting a dual RAG MAS
Phase 1: Data preparation
Phase 2: Runtime execution analysis
RAG pipeline data ingestion (context and knowledge)
Installation and setup
Initializing the Pinecone index
Data preparation: the context library (procedural RAG)
Data preparation: the knowledge base (factual RAG)
Process and upload (upsert) data
Building the context-aware system
Summary
Questions
References
Further reading
4 Assembling the Context Engine
Architecting the Context Engine
Architectural overview
Functional deep dive
Assembling the system
The specialist agents
Agent Registry
The Context Engine
Putting it all together
Running the engine
Summary
Questions
References
Further reading
5 Hardening the Context Engine
Step-by-step architectural walk-through
Phase 1: Initiation
Phase 2: The planning stage
Phase 3: The execution loop
Phase 4: Finalization
Refactoring for scalability in the real world
Creating the centralized setup functions
Simplifying the main notebook
Hardening the helper functions
Refactoring the agents for production
Refactoring the Agent Registry
Upgrading the central orchestrator
Running the hardened engine
Deconstructing the engine's thought process
Finalization
Modularizing the engine
Local imports
Module independence
The final pre-production notebook
Summary
Questions
References
Further reading
6 Building the Summarizer Agent for Context Reduction
Architecting the glass-box system
Step-by-step architectural walkthrough
Separation of responsibilities
Why the glass box matters
Implementing context reduction with the Summarizer agent
Foundation for cost management
Building the Summarizer agent
Integrating the new agent into the engine's toolkit
Reinforcing the Writer agent for greater flexibility
Exploring the Summarizer in action
Translating technical efficiency into business value
Summary
Questions
References
Further reading
7 High-Fidelity RAG and Defense: The NASA-Inspired Research Assistant
Architecting a trustworthy research assistant
Step-by-step architectural walkthrough
Separation of responsibilities
Implementing high-fidelity RAG and agent defenses
Part 1: Upgrading the ingestion pipeline
Part 2: Upgrading the context engine's capabilities
Part 3: The final application: the NASA research assistant
Validation and retro compatibility of the context engine
Complete inventory of the context engine
How the context engine thinks
Validating the mind of the machine
Summary
Questions
References
Further reading
8 Architecting for Reality: Moderation, Latency, and Policy-Driven AI
Architecting the enterprise-ready Context Engine
Step-by-step architectural walkthrough
The deliberate pace of a reasoning engine
Implementing moderation
Building the moderation gatekeeper
Integrating the moderation gatekeeper
Moderation guardrail in action
Architecting a policy-driven meta-controller
Principle 1: AI systems must continuously adapt to reality
Principle 2: Limits of automated contextual judgment
Principle 3: New engineer's mindset
Principle 4: Policy as the ultimate context
Principle 5: Architectural solution
Multi-domain, generic control decks
Template 1: High-fidelity RAG
Template 2: Context reduction
Template 3: Grounded reasoning
Applying the engine: the legal compliance assistant
Building the legal knowledge base
Legal use cases and real-world limits
Summary
Questions
References
Further reading
9 Architecting for Brand and Agility: The Strategic Marketing Engine
Step-by-step architectural flow
Designing the marketing knowledge base
Running the engine
Validating production safeguards
Use case 1: Competitive analysis
Use case 2: Transforming technical specs into marketing copy
Use case 3: Synthesizing a persuasive pitch from multiple sources
Summary
Questions
References
Further reading
10 The Blueprint for Production-Ready AI
Productionizing the glass-box engine
Environment configuration and secrets management
Building the production API (orchestration layer)
Asynchronous execution and task queues
Centralized logging and observability
Infrastructure and containerization
Deploying enterprise capabilities and production guardrails
Managing operational costs with proactive context reduction
Ensuring trust and compliance with high-fidelity RAG
Defending the data pipeline against poisoning and adversarial attacks
Ensuring compliance and safety with automated guardrails
Enforcing governance and quality with creative workflows
Presenting the business value
From cost center to value multiplier
Stakeholder trust through verifiability and security
Creating a strategic asset
Summary
Questions
References
Further reading
11 Unlock Your Exclusive Benefits
Unlock this Book’s Free Benefits in 3 Easy Steps
Appendix A Context Engine Reference Guide
Theoretical foundations
Semantic blueprint
Merits of the glass-box architecture
System architecture and workflow
The commons library reference
Data ingestion pipeline
Pipeline steps
Ingesting the context library
Execution and operation
Engine room
Control deck
Production safeguards: moderation, sanitization, and policy
Limits of automation and the role of policy
Operational realities: latency and stochasticity
Appendix B Answers
Chapter 1
Chapter 2
Chapter 3
Chapter 4
Chapter 5
Chapter 6
Chapter 7
Chapter 8
Chapter 9
Chapter 10
Other Books You May Enjoy
Index

简介

版权

Context Engineering for Multi-Agent Systems