English manuscript: reading order¶
Start with the title page, then read the preface, reading guide, nine parts, and closing matter. Chapter numbering restarts in each part. The list follows chapter numbers within each part, not the website's module order. These links use the same chapter sources; there is no separate copy of the prose.
Part 1: Large Language Models (23 chapters)¶
- Part 1, Chapter 1: How Large Language Models Differ from Traditional NLP
- Part 1, Chapter 2: Transformer Architecture
- Part 1, Chapter 3: MHA Limitations, MQA, GQA, and FlashAttention
- Part 1, Chapter 4: Positional Encoding
- Part 1, Chapter 5: Tokenizers
- Part 1, Chapter 6: The Three-Stage Training Framework for LLMs
- Part 1, Chapter 7: Scaling Laws and Emergent Abilities
- Part 1, Chapter 8: Fine-Tuning Approaches for LLMs
- Part 1, Chapter 9: LoRA in Depth
- Part 1, Chapter 10: A Survey of Post-Training Methods
- Part 1, Chapter 11: DPO vs. PPO in Depth
- Part 1, Chapter 12: Decoding Strategies
- Part 1, Chapter 13: Tuning Temperature, Top-P, and Top-K
- Part 1, Chapter 14: KV Cache and Prompt Caching
- Part 1, Chapter 15: Model Quantization
- Part 1, Chapter 16: Prompt Engineering
- Part 1, Chapter 17: Chain of Thought
- Part 1, Chapter 18: Hallucinations—Causes and Mitigation
- Part 1, Chapter 19: Mixture-of-Experts Models
- Part 1, Chapter 20: Choosing a Serving Framework
- Part 1, Chapter 21: Capability Evaluation Metrics
- Part 1, Chapter 22: Practical Model Selection
- Part 1, Chapter 23: Multimodal Models
Part 2: Multimodal AI (10 chapters)¶
- Part 2, Chapter 1: Multimodal Representations and Fusion Architectures
- Part 2, Chapter 2: Vision–Language Models and Visual Grounding
- Part 2, Chapter 3: OCR and Document AI
- Part 2, Chapter 4: Computer Use and GUI Agents
- Part 2, Chapter 5: Speech Recognition, Synthesis, and Audio Language Models
- Part 2, Chapter 6: Real-Time Full-Duplex Voice Interaction
- Part 2, Chapter 7: Diffusion Models, Flow Matching, and Image Generation
- Part 2, Chapter 8: Video Generation Models
- Part 2, Chapter 9: Multimodal Training Data and Alignment
- Part 2, Chapter 10: Multimodal Evaluation, Safety, and Inference Serving
Part 3: Tools and Protocols (15 chapters)¶
- Part 3, Chapter 1: What Function Calling Is and How It Works
- Part 3, Chapter 2: How LLMs Learn to Use Tools
- Part 3, Chapter 3: Tool Definitions and Schema Engineering
- Part 3, Chapter 4: The Core of the Model Context Protocol
- Part 3, Chapter 5: The Three Layers of MCP
- Part 3, Chapter 6: MCP and Function Calling—Differences and Tradeoffs
- Part 3, Chapter 7: Why Some Reasoning Models Seem Not to Support MCP
- Part 3, Chapter 8: What Is a Skill?
- Part 3, Chapter 9: Skills and MCP
- Part 3, Chapter 10: How Function Calling, MCP, and Skills Fit Together
- Part 3, Chapter 11: The A2A Protocol
- Part 3, Chapter 12: MCP Transports
- Part 3, Chapter 13: SSE, WebSocket, and WebRTC
- Part 3, Chapter 14: LLM Gateways
- Part 3, Chapter 15: Tool Protocol Security
Part 4: Retrieval-Augmented Generation (22 chapters)¶
- Part 4, Chapter 1: What RAG Is and What Problems It Solves
- Part 4, Chapter 2: Tradeoffs Between RAG, Fine-Tuning, and Long Context
- Part 4, Chapter 3: Document Parsing and Preprocessing
- Part 4, Chapter 4: Chunking Strategies and Granularity
- Part 4, Chapter 5: When Chunking Breaks the Meaning
- Part 4, Chapter 6: Embedding Principles and Technical Evolution
- Part 4, Chapter 7: Selecting and Evaluating Embedding Models
- Part 4, Chapter 8: Vector Databases and ANN Indexes
- Part 4, Chapter 9: Vector Databases in Production and Performance Tuning
- Part 4, Chapter 10: The Complete Online RAG Pipeline
- Part 4, Chapter 11: Retrieval Paradigms—Sparse, Dense, and Late Interaction
- Part 4, Chapter 12: Query Understanding and Rewriting
- Part 4, Chapter 13: Multi-Path Retrieval, RRF Fusion, and Reranking
- Part 4, Chapter 14: A Five-Layer Framework for RAG Optimization
- Part 4, Chapter 15: Advanced RAG Paradigms
- Part 4, Chapter 16: GraphRAG and Graph Retrieval
- Part 4, Chapter 17: Generation, Grounding, and Hallucination Mitigation
- Part 4, Chapter 18: A Framework for Evaluating RAG
- Part 4, Chapter 19: Dynamic Knowledge Base Updates and Incremental Indexing
- Part 4, Chapter 20: RAG Deployment Challenges and Security
- Part 4, Chapter 21: Multimodal RAG
- Part 4, Chapter 22: Text-to-SQL: From Business Questions to Verifiable Queries
Part 5: Agents (25 chapters)¶
- Part 5, Chapter 1: From Large Language Models to AI Agents
- Part 5, Chapter 2: Modern Agent System Architecture
- Part 5, Chapter 3: Tools, Skills, Agents, Workflows, and AGENTS.md
- Part 5, Chapter 4: Agent Design Patterns
- Part 5, Chapter 5: Model Reasoning and Search Methods for Agents
- Part 5, Chapter 6: Decomposing and Scheduling Complex Tasks
- Part 5, Chapter 7: Memory Mechanisms in AI Agents
- Part 5, Chapter 8: Implementing Short- and Long-Term Agent Memory
- Part 5, Chapter 9: Single-Agent and Multi-Agent Systems
- Part 5, Chapter 10: Agent Memory and Context Compression
- Part 5, Chapter 11: Giving LLMs and Agents the Ability to Plan
- Part 5, Chapter 12: Agent Reflection, Verification, and Self-improvement
- Part 5, Chapter 13: Multi-Agent Coordination, Routing, and Dynamic Switching
- Part 5, Chapter 14: Agent Evaluation and Benchmarks
- Part 5, Chapter 15: Agent Security and Prompt Injection
- Part 5, Chapter 16: Agent Harnesses—Definitions, Boundaries, and Layers
- Part 5, Chapter 17: The Agent Loop and Runtime State Machine
- Part 5, Chapter 18: Context Engineering and Context Assembly
- Part 5, Chapter 19: Tool Registries, Call Contracts, and Execution Pipelines
- Part 5, Chapter 20: Permissions, Sandboxing, and Isolation
- Part 5, Chapter 21: Checkpoints, Persistence, Retries, Timeouts, Idempotency, and Recovery
- Part 5, Chapter 22: Human-in-the-Loop Approval and Interruptible Execution
- Part 5, Chapter 23: Tracing, Evaluation, Cost Control, and Coding-Agent Harness Case Studies
- Part 5, Chapter 24: Code Search, Editing, and Verification for Coding Agents
- Part 5, Chapter 25: Agent Post-Training: From Failed Trajectories to Reliable Policies
Part 6: Frameworks and Orchestration (23 chapters)¶
- Part 6, Chapter 1: An Overview of Major AI Agent Development Frameworks
- Part 6, Chapter 2: Chain Design and LCEL
- Part 6, Chapter 3: The Underlying Architecture of LangChain v1
- Part 6, Chapter 4: Seven Steps to a Production-Ready LangChain Agent
- Part 6, Chapter 5: Tool Registration and Tool Contracts
- Part 6, Chapter 6: Short-Term and Long-Term Memory in LangChain
- Part 6, Chapter 7: Dividing Responsibilities Between LangChain and LlamaIndex
- Part 6, Chapter 8: LangChain4j and the Java Ecosystem
- Part 6, Chapter 9: The Layered Relationship Between LangChain and LangGraph
- Part 6, Chapter 10: LangGraph's Core Advantages
- Part 6, Chapter 11: LangChain's Version Evolution
- Part 6, Chapter 12: How Deep Research Works
- Part 6, Chapter 13: Building a Production Quality Feedback Loop with LangSmith
- Part 6, Chapter 14: LlamaIndex Data and Index Abstractions
- Part 6, Chapter 15: LlamaIndex Query Engines and Workflow Orchestration
- Part 6, Chapter 16: DSPy's Declarative Programming Model: Signatures, Modules, and Programs
- Part 6, Chapter 17: DSPy's Compiler and Optimizers
- Part 6, Chapter 18: Semantic Kernel's Core Abstractions: Kernel, Plugin, and Planner
- Part 6, Chapter 19: Semantic Kernel's Process Framework and Agent Framework
- Part 6, Chapter 20: Multi-Agent Abstractions in AutoGen and CrewAI
- Part 6, Chapter 21: PydanticAI's Type-Safe Approach and Where the Three Frameworks Fit
- Part 6, Chapter 22: Comparing Framework Internals: State, Persistence, Tool Contracts, and Observability
- Part 6, Chapter 23: Identifying Lock-in, Designing Portable Architectures, and Planning Migrations
Part 7: Production Engineering (13 chapters)¶
- Part 7, Chapter 1: What Is LLMOps? How It Relates to MLOps and DevOps
- Part 7, Chapter 2: A Production Architecture for AI Applications
- Part 7, Chapter 3: Model Gateways, Routing, and Fallback
- Part 7, Chapter 4: Retries, Timeouts, Idempotency, and Circuit Breakers
- Part 7, Chapter 5: Structured Outputs and Contract Validation
- Part 7, Chapter 6: Guardrails and Graceful Degradation
- Part 7, Chapter 7: Offline Evaluation and Eval-Driven Development
- Part 7, Chapter 8: Production Observability and Tracing
- Part 7, Chapter 9: Prompt, Model, and Data Versioning
- Part 7, Chapter 10: LLM CI/CD, Staged Rollouts, Canary Releases, and A/B Tests
- Part 7, Chapter 11: Caching, Batching, Throughput, and Cost Optimization
- Part 7, Chapter 12: SLOs, Capacity Planning, and Incident Response
- Part 7, Chapter 13: Feedback Loops and the Data Flywheel
Part 8: Safety and Governance (10 chapters)¶
- Part 8, Chapter 1: AI System Threat Modeling and the Attack Surface
- Part 8, Chapter 2: Prompt Injection, Jailbreaks, and System Prompt Leakage
- Part 8, Chapter 3: Output Handling and Secret Exfiltration
- Part 8, Chapter 4: Training, Fine-Tuning, and RAG Data Poisoning
- Part 8, Chapter 5: Model Supply Chains and Serialization Risks
- Part 8, Chapter 6: Privacy, PII, and Memory Leakage
- Part 8, Chapter 7: Least Privilege and Identity Governance for Agents, Tools, MCP, and A2A
- Part 8, Chapter 8: Sandboxing and Network Isolation for Code Execution, Browsers, and Computer Use
- Part 8, Chapter 9: Security Evaluation, Red Teaming, and Continuous Assurance
- Part 8, Chapter 10: AI Governance, Risk Classification, Auditing, and Content Provenance
Part 9: Forward Deployed Engineering (2 chapters)¶
- Part 9, Chapter 1: Forward Deployed Engineering: From Business Problems to Reusable Production Systems
- Part 9, Chapter 2: FDE Delivery Pitfalls and Lessons