Skip to content

References and Further Reading

Both language editions use the same reference numbers. Superscript 【numbers】 in the text link to the corresponding source; numbers identify sources, not their importance. The original chapter-end reading suggestions and source qualifications remain under “Chapter reading notes” below. A source listed only as further reading is not presented as direct evidence for an individual paragraph.

Numbered sources

【1】 BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

【2】 Language Models are Few-Shot Learners (GPT-3)

【3】 Language Models are Unsupervised Multitask Learners (GPT-2)

【4】 Emergent Abilities of Large Language Models

【5】 Are Emergent Abilities of Large Language Models a Mirage?

【6】 Scaling Laws for Neural Language Models

【7】 Training Compute-Optimal Large Language Models (Chinchilla)

【8】 The Llama 3 Herd of Models

【9】 Attention Is All You Need

【10】 Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer (T5)

【11】 On Layer Normalization in the Transformer Architecture

【12】 Root Mean Square Layer Normalization

【13】 GLU Variants Improve Transformer

【14】 Query-Key Normalization for Transformers

【15】 Transformer Feed-Forward Layers Are Key-Value Memories

【16】 Fast Transformer Decoding: One Write-Head is All You Need (MQA)

【17】 GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints

【18】 Llama 2

【19】 DeepSeek-V2 paper and official implementation

【20】 FlashAttention

【21】 FlashAttention official repository and version requirements (README snapshot, 2026-07-06)

【22】 Online normalizer calculation for softmax

【23】 Mistral 7B

【24】 Linformer

【25】 Rethinking Attention with Performers

【26】 Mamba: Linear-Time Sequence Modeling with Selective State Spaces

【27】 RoFormer: Enhanced Transformer with Rotary Position Embedding

【28】 ALiBi: Train Short, Test Long

【29】 Position Interpolation

【30】 YaRN: Efficient Context Window Extension of Large Language Models

【31】 Hugging Face Transformers: RoPE parameters and variants

【32】 BLOOM model card

【33】 MosaicML: MPT-7B release announcement and ALiBi configuration

【34】 Lost in the Middle: How Language Models Use Long Contexts

【35】 Neural Machine Translation of Rare Words with Subword Units (BPE)

【36】 SentencePiece: A simple and language independent subword tokenizer

【37】 Subword Regularization: Improving NMT Models with Multiple Subword Candidates (Unigram)

【38】 Hugging Face: Tokenizers tutorial

【39】 OpenAI tiktoken

【40】 Meta Llama 3 tokenizer implementation

【41】 Google SentencePiece implementation and configuration

【42】 Hugging Face Transformers: Tokenization algorithms

【43】 Hugging Face Transformers: Chat templates

【44】 GPT-3, §2 and few-shot evaluations

【45】 Llama 2, §§2.1 and 3.1.1, and the model card

【46】 The Llama 3 Herd of Models, §§3–4

【47】 Chinchilla: the training-compute approximation

【48】 InstructGPT

【49】 LIMA

【50】 Deep Reinforcement Learning from Human Preferences

【51】 DPO: reward reparameterization and preference loss

【52】 DeepSeek-R1: the training workflow in the initial 2025 report

【53】 Qwen3 Technical Report, initial version

【54】 DeepSeekMath: data construction and GRPO

【55】 LoRA

【56】 QLoRA

【57】 Transformers v4.46.3: Model training anatomy

【58】 PEFT: LoRA configuration and initialization

【59】 Parameter-Efficient Transfer Learning for NLP

【60】 PEFT v0.17.0: LoRA configuration, initialization, and scaling

【61】 PEFT: Model merging and add_weighted_adapter

【62】 PEFT v0.17.0: weighted-adapter implementation and limitations

【63】 vLLM: LoRA adapters

【64】 Constitutional AI

【65】 RLAIF vs. RLHF

【66】 Proximal Policy Optimization Algorithms

【67】 OpenAI Spinning Up: PPO-Clip equations and constraint limits

【68】 High-Dimensional Continuous Control Using Generalized Advantage Estimation

【69】 The Curious Case of Neural Text Degeneration

【70】 If Beam Search Is the Answer, What Was the Question?

【71】 Fast Inference from Transformers via Speculative Decoding

【72】 Self-Consistency Improves Chain of Thought Reasoning in Language Models

【73】 PagedAttention: cache sharing, including beam search

【74】 Qwen3-30B-A3B official model card

【75】 vLLM: Batch Invariance

【76】 Transformers: Generation Utilities, including Temperature/TopK/TopP LogitsWarper

【77】 Transformers v4.56.2: configuration conditions for greedy decoding, sampling, and beam search

【78】 Anthropic official prompt caching cookbook

【79】 Anthropic official SDK: cache TTL type

【80】 OpenAI: Prompt Caching

【81】 KIVI paper

【82】 vLLM: Automatic Prefix Caching design

【83】 GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers

【84】 AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration

【85】 LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale

【86】 SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models

【87】 Optimal Brain Compression: A Framework for Accurate Post-Training Quantization and Pruning

【88】 llama.cpp / GGUF

【89】 AutoAWQ

【90】 GPTQ authors' implementation: column order, act-order, and grouping

【91】 AWQ paper: per-channel scaling and search

【92】 PyTorch AO: Quantization-Aware Training

【93】 FP8 Formats for Deep Learning, v1, §§2–3 and Table 1

【94】 OCP Microscaling Formats (MX) Specification v1.0

【95】 Microsoft MX library: max-exponent scaling and scale limits, pinned revision

【96】 NVIDIA: Introducing NVFP4 for Efficient and Accurate Low-Precision Inference

【97】 NVIDIA Transformer Engine: FP8 primer and MXFP8/NVFP4, pinned notebook

【98】 NVIDIA Transformer Engine: scaling recipes, history, and margin, pinned source

【99】 gpt-oss-120b & gpt-oss-20b model card, v1, §2.1 and Table 1: MXFP4 MoE weights and checkpoint sizes

【100】 AMD ROCm blog: FP8 FNUZ and OCP variants on CDNA3 and CDNA4

【101】 ONNX: FP8 normal/subnormal decoding and special-value encodings

【102】 vLLM v0.10.2: E4M3FN-to-E4M3FNUZ normalization, pinned implementation

【103】 Calibrate Before Use: Improving Few-Shot Performance of Language Models (example and ordering biases)

【104】 Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

【105】 Large Language Models are Zero-Shot Reasoners (Let's think step by step)

【106】 LLMLingua: Compressing Prompts for Accelerated Inference of Large Language Models

【107】 LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compression

【108】 The Power of Scale for Parameter-Efficient Prompt Tuning (soft prompts)

【109】 OpenAI: Reasoning best practices (including the scope of delimiters and CoT prompts)

【110】 OpenAI: Structured Outputs

【111】 Least-to-Most Prompting Enables Complex Reasoning in Large Language Models

【112】 Tree of Thoughts: Deliberate Problem Solving with Large Language Models

【113】 Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matters

【114】 Measuring Faithfulness in Chain-of-Thought Reasoning

【115】 s1: Simple test-time scaling

【116】 Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs

【117】 OpenAI: Reasoning models

【118】 Anthropic Python SDK: manual thinking configuration

【119】 Amazon Bedrock: Extended thinking

【120】 On Faithfulness and Factuality in Abstractive Summarization

【121】 Survey of Hallucination in Natural Language Generation

【122】 A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions

【123】 Language Models (Mostly) Know What They Know

【124】 TruthfulQA: Measuring How Models Mimic Human Falsehoods

【125】 Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

【126】 SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models

【127】 Chain-of-Verification Reduces Hallucination in Large Language Models

【128】 Towards Understanding Sycophancy in Language Models

【129】 Sparsely-Gated Mixture-of-Experts Layer

【130】 GShard

【131】 Switch Transformers

【132】 Mixtral of Experts

【133】 DeepSeek-V3 Technical Report

【134】 Auxiliary-Loss-Free Load Balancing

【135】 Expert Choice Routing

【136】 ST-MoE: router z-loss

【137】 SGLang: Efficient Execution of Structured Language Model Programs (RadixAttention)

【138】 Orca: A Distributed Serving System for Transformer-Based Generative Models (continuous batching)

【139】 vLLM documentation

【140】 vLLM: Automatic Prefix Caching

【141】 SGLang repository

【142】 Text Generation Inference repository

【143】 TGI official documentation: maintenance-mode notice, checked in the source manuscript on 2026-09-15

【144】 TensorRT-LLM repository

【145】 llama.cpp: HTTP server capabilities and parameters

【146】 TensorRT-LLM: Quick Start

【147】 vLLM: Chunked Prefill and configuration tuning

【148】 vLLM: Disaggregated Prefilling

【149】 TGI: components and HTTP/gRPC boundaries

【150】 TensorRT-LLM: LLM API and PyTorch backend

【151】 Measuring Massive Multitask Language Understanding (MMLU)

【152】 MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark

【153】 Evaluating Large Language Models Trained on Code (HumanEval / Pass@k)

【154】 SWE-bench: Can Language Models Resolve Real-World GitHub Issues?

【155】 OpenAI: Why SWE-bench Verified no longer measures frontier coding capabilities

【156】 Training Verifiers to Solve Math Word Problems (GSM8K)

【157】 Measuring Mathematical Problem Solving With the MATH Dataset

【158】 GPQA: A Graduate-Level Google-Proof Q&A Benchmark

【159】 Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena

【160】 Holistic Evaluation of Language Models (HELM)

【161】 LiveBench: A Challenging, Contamination-Limited LLM Benchmark

【162】 tau-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains

【163】 HumanEval's official Pass@k implementation

【164】 Official SWE-bench documentation: Verified subset

【165】 Humanity's Last Exam

【166】 DeepSeek-V3 Technical Report

【167】 Qwen3 Technical Report

【168】 RouteLLM: Learning to Route LLMs with Preference Data

【169】 FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance

【170】 OpenAI: Data controls in the OpenAI platform

【171】 CLIP: Learning Transferable Visual Models From Natural Language Supervision

【172】 Flamingo: a Visual Language Model for Few-Shot Learning

【173】 LLaVA: Visual Instruction Tuning

【174】 Whisper: Robust Speech Recognition via Large-Scale Weak Supervision

【175】 Video-LLaMA: An Instruction-tuned Audio-Visual Language Model

【176】 NIST AI Risk Management Framework

【177】 BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models

【178】 Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution

【179】 Hugging Face Evaluate: WER definition and implementation notes

【180】 Sigmoid Loss for Language Image Pre-Training (SigLIP)

【181】 ImageBind: One Embedding Space To Bind Them All

【182】 Perceiver IO: A General Architecture for Structured Inputs & Outputs

【183】 Qwen-VL: Position-aware vision–language adapter

【184】 Chameleon: Mixed-Modal Early-Fusion Foundation Models

【185】 Neural Discrete Representation Learning (VQ-VAE)

【186】 Taming Transformers for High-Resolution Image Synthesis (VQGAN)

【187】 OpenAI GPT-4o System Card

【188】 GPT-4o System Card (original report)

【189】 Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection

【190】 Grounded Language-Image Pre-training (GLIP)

【191】 Simple Open-Vocabulary Object Detection with Vision Transformers (OWL-ViT)

【192】 PaliGemma: A versatile 3B VLM for transfer

【193】 Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Multimodal Models

【194】 Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations

【195】 Generation and Comprehension of Unambiguous Object Descriptions (RefCOCOg)

【196】 ReferItGame: Referring to Objects in Photographs of Natural Scenes

【197】 REFER: Dataset origins and split documentation

【198】 LayoutLM: Pre-training of Text and Layout for Document Image Understanding

【199】 LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking

【200】 OCR-free Document Understanding Transformer (Donut)

【201】 PubTables-1M: Towards Comprehensive Table Extraction From Unstructured Documents

【202】 GriTS: Grid Table Similarity

【203】 Image-based table recognition: data, model, and evaluation (PubTabNet / TEDS)

【204】 ChartQA: A Benchmark for Question Answering about Charts

【205】 DocVQA: A Dataset for VQA on Document Images

【206】 Azure AI Document Intelligence documentation

【207】 Azure Document Intelligence: Accuracy and confidence

【208】 Google Cloud Document AI documentation

【209】 Amazon Textract documentation

【210】 Amazon Textract: AnalyzeDocument and FeatureTypes

【211】 Anthropic: Developing a computer use model

【212】 Anthropic: Upgraded Claude 3.5 Sonnet and computer use public beta

【213】 Anthropic Computer Use documentation

【214】 OpenAI: Computer-Using Agent

【215】 OpenAI Operator

【216】 OpenAI: Computer use tools and execution environments

【217】 UI-TARS: Pioneering Automated GUI Interaction with Native Agents

【218】 OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments

【219】 WebArena: A Realistic Web Environment for Building Autonomous Agents

【220】 Connectionist Temporal Classification (original paper)

【221】 Sequence Transduction with Recurrent Neural Networks (RNN-T)

【222】 Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions (Tacotron 2)

【223】 Neural Codec Language Models are Zero-Shot Text to Speech Synthesizers (VALL-E)

【224】 SoundStream: An End-to-End Neural Audio Codec

【225】 High Fidelity Neural Audio Compression (EnCodec)

【226】 AudioLM: a Language Modeling Approach to Audio Generation

【227】 Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models

【228】 SALMONN: Towards Generic Hearing Abilities for Large Language Models

【229】 Moshi: a speech-text foundation model for real-time dialogue

【230】 Generative Spoken Dialogue Language Modeling (dGSLM)

【231】 OpenAI Realtime API documentation

【232】 OpenAI Realtime: Conversation state and interruption handling

【233】 Kyutai Moshi project page

【234】 Denoising Diffusion Probabilistic Models (DDPM)

【235】 Denoising Diffusion Implicit Models (DDIM)

【236】 Score-Based Generative Modeling through Stochastic Differential Equations

【237】 High-Resolution Image Synthesis with Latent Diffusion Models (Stable Diffusion)

【238】 Classifier-Free Diffusion Guidance

【239】 Flow Matching for Generative Modeling

【240】 Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow

【241】 Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (Stable Diffusion 3)

【242】 Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding (Imagen)

【243】 OpenAI DALL·E 3: Improving Image Generation with Better Captions

【244】 GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium (FID)

【245】 CLIPScore: A Reference-free Evaluation Metric for Image Captioning

【246】 Video Diffusion Models

【247】 Imagen Video: High Definition Video Generation with Diffusion Models

【248】 OpenAI: Video generation models as world simulators (Sora technical report)

【249】 Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets

【250】 Towards Accurate Generative Models of Video: A New Metric & Challenges (FVD)

【251】 VBench: Comprehensive Benchmark Suite for Video Generative Models

【252】 LAION-5B: An Open Large-Scale Dataset for Training Next Generation Image-Text Models

【253】 OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents

【254】 Aligning Large Multimodal Models with Factually Augmented RLHF (LLaVA-RLHF)

【255】 MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark

【256】 MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts

【257】 Multimodal Neurons in Artificial Neural Networks (Typographic Attack)

【258】 Visual Adversarial Examples Jailbreak Aligned Large Language Models

【259】 C2PA 2.1 technical specification

【260】 Google DeepMind: SynthID

【261】 DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving

【262】 OpenAI: Function Calling guide (checked 2026-09-08)

【263】 OpenAI: Responses API migration guide (this section's fields and examples were checked against this page and the Function Calling guide on 2026-09-16)

【264】 OpenAI: The 2023 Function Calling announcement

【265】 Anthropic: Tool Use with Claude

【266】 Anthropic: Writing Effective Tools for Agents

【267】 Toolformer: Language Models Can Teach Themselves to Use Tools

【268】 ReAct: Synergizing Reasoning and Acting in Language Models

【269】 JSON Schema specification

【270】 ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs

【271】 ToolRL: Reward is All Tool Learning Needs

【272】 ReTool: Reinforcement Learning for Strategic Tool Use in LLMs

【273】 Berkeley Function Calling Leaderboard

【274】 API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs

【275】 Anthropic: Building Effective Agents

【276】 MCP version status

【277】 MCP 2026-07-28 changelog

【278】 Model Context Protocol documentation

【279】 MCP specification 2026-07-28

【280】 MCP version compatibility

【281】 MCP architecture

【282】 Anthropic: Introducing the Model Context Protocol

【283】 JSON-RPC 2.0 specification

【284】 github/github-mcp-server

【285】 MCP Server capabilities: Tools

【286】 MCP Multi Round-Trip Requests

【287】 MCP deprecated features

【288】 MCP transport specification

【289】 subscription specification

【290】 MCP Server development quickstart

【291】 MCP Servers official examples

【292】 Archived MCP examples

【293】 MCP Roots: deprecation and the absence of a security boundary

【294】 OpenAI: Function calling guide

【295】 OpenAI: o1-preview model page

【296】 OpenAI: o1 model and snapshots

【297】 OpenAI Cookbook: Early o1-preview limitations on structured outputs

【298】 Anthropic: Claude 4 announcement of interleaved tool use

【299】 Anthropic Cookbook: Early Claude 3.7 example of thinking-block preservation

【300】 Anthropic: Extended Thinking

【301】 DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

【302】 Anthropic: Introducing Agent Skills

【303】 Anthropic: Equipping Agents for the Real World with Agent Skills

【304】 Agent Skills specification

【305】 Agent Skills pinned commit used for the format review

【306】 Claude Docs: Agent Skills

【307】 Anthropic: Effective Context Engineering for AI Agents

【308】 official client list

【309】 A2A v1.0.1 release notes

【310】 A2A protocol website

【311】 A2A v1.0.1 release specification

【312】 A2A v1.0.1 protocol data definitions

【313】 Google: Announcing the Agent2Agent Protocol

【314】 Linux Foundation: A2A Project

【315】 MCP 2026-07-28 Streamable HTTP

【316】 MCP 2026-07-28 stdio

【317】 MCP specification 2025-03-26: the introduction of Streamable HTTP

【318】 MDN: Server-Sent Events

【319】 MDN: The WebSocket API

【320】 MDN: WebRTC API

【321】 RFC 6455: The WebSocket Protocol

【322】 RFC 8445: Interactive Connectivity Establishment (ICE)

【323】 WebRTC Official Site

【324】 WHATWG HTML: Server-Sent Events

【325】 RFC 8835: WebRTC Transports

【326】 RFC 8834: WebRTC Media and Packet-Loss Recovery

【327】 RFC 8831: WebRTC Data Channels

【328】 RFC 8441: WebSocket over HTTP/2

【329】 RFC 9220: WebSocket over HTTP/3

【330】 OpenAI: WebRTC Connections (Realtime API Section)

【331】 Squid v5 supports controlled upgrades

【332】 LiteLLM Documentation

【333】 LiteLLM GitHub

【334】 Bifrost Official Repository (Go Implementation)

【335】 LiteLLM Retries and Failover

【336】 RFC 9111: HTTP Caching

【337】 Portkey AI Gateway

【338】 Kong AI Gateway

【339】 Envoy AI Gateway

【340】 Langfuse: LLM Observability

【341】 OWASP Top 10 for LLM Applications

【342】 MCP 2026-07-28 Authorization

【343】 MCP 2026-07-28 Security Best Practices

【344】 RFC 9728: OAuth 2.0 Protected Resource Metadata

【345】 RFC 8252: OAuth for Native Apps and loopback redirects

【346】 RFC 9700: OAuth 2.0 Security Best Current Practice

【347】 RFC 7636: PKCE

【348】 RFC 8707: Resource Indicators for OAuth 2.0

【349】 Dense Passage Retrieval for Open-Domain Question Answering

【350】 Retrieval-Augmented Generation for Large Language Models: A Survey

【351】 Searching for Best Practices in Retrieval-Augmented Generation

【352】 Anthropic: Introducing Contextual Retrieval

【353】 Chroma Research: Context Rot — How Increasing Input Tokens Impacts LLM Performance

【354】 Can Long-Context Language Models Subsume Retrieval, RAG, SQL, and More?

【355】 Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG

【356】 LongRAG: Enhancing Retrieval-Augmented Generation with Long-context LLMs

【357】 RAFT: Adapting Language Model to Domain Specific RAG

【358】 ColPali: Efficient Document Retrieval with Vision Language Models

【359】 Qdrant: Payload Updates and Indexing

【360】 Advanced ingestion process powered by LLM parsing for RAG system

【361】 Dense X Retrieval: What Retrieval Granularity Should We Use?

【362】 Qwen3-Embedding-0.6B Model Card

【363】 Late Chunking: Contextual Chunk Embeddings Using Long-Context Embedding Models

【364】 Jina AI: Post-Encoding Pooling and Boundaries in Late Chunking

【365】 RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval

【366】 LangChain's ParentDocumentRetriever

【367】 Efficient Estimation of Word Representations in Vector Space

【368】 Stanford GloVe: global co-occurrence statistics and a weighted least-squares objective

【369】 Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks

【370】 SimCSE: Simple Contrastive Learning of Sentence Embeddings

【371】 ColBERTv2: Effective and Efficient Retrieval via Lightweight Late Interaction

【372】 MTEB: Massive Text Embedding Benchmark

【373】 MMTEB: Massive Multilingual Text Embedding Benchmark

【374】 Matryoshka Representation Learning

【375】 OpenAI: New embedding models and API updates (2024-01-25)

【376】 OpenAI: Embedding API and the dimensions parameter

【377】 M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation

【378】 Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs

【379】 pgvector official README: filtering, iterative scans, VACUUM, and scaling

【380】 FreshDiskANN: A Fast and Accurate Graph-Based ANN Index for Streaming Similarity Search

【381】 ACORN: Performant and Predicate-Agnostic Search Over Vector Embeddings and Structured Data

【382】 Billion-scale similarity search with GPUs

【383】 hnswlib v0.8.0: internal IDs, base-layer adjacency capacity, and payload layout

【384】 Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection

【385】 Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity

【386】 Lucene 9.12 BM25Similarity: term-frequency saturation, length normalization, and IDF implementation

【387】 SPLADE: Sparse Lexical and Expansion Model for First Stage Ranking

【388】 Elastic: official ELSER documentation

【389】 Precise Zero-Shot Dense Retrieval without Relevance Labels

【390】 Take a Step Back: Evoking Reasoning via Abstraction in Large Language Models

【391】 Query Rewriting for Retrieval-Augmented Large Language Models

【392】 Reciprocal Rank Fusion outperforms Condorcet and individual Rank Learning Methods (SIGIR 2009)

【393】 Evaluation of Retrieval-Augmented Generation: A Survey

【394】 Corrective Retrieval Augmented Generation

【395】 Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG

【396】 Active Retrieval Augmented Generation

【397】 From Local to Global: A Graph RAG Approach to Query-Focused Summarization

【398】 Microsoft GraphRAG documentation

【399】 Microsoft GraphRAG: Indexing Methods

【400】 Microsoft GraphRAG: CLI Reference, including update methods

【401】 Microsoft GraphRAG: Query Modes

【402】 Microsoft GraphRAG: Local Search

【403】 LightRAG: Simple and Fast Retrieval-Augmented Generation

【404】 HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models

【405】 PathRAG: Pruning Graph-based Retrieval Augmented Generation with Relational Paths

【406】 Astute RAG: Overcoming Imperfect Retrieval Augmentation and Knowledge Conflicts for Large Language Models

【407】 Making Retrieval-Augmented Language Models Robust to Irrelevant Context

【408】 RAGAS: Automated Evaluation of Retrieval Augmented Generation

【409】 ALCE: Enabling Large Language Models to Generate Text with Citations

【410】 Ragas: Context Recall

【411】 Ragas: Context Precision

【412】 Ragas: Faithfulness

【413】 Ragas: Answer / Response Relevancy

【414】 CRAG - Comprehensive RAG Benchmark

【415】 TREC RAG Track

【416】 Fact, Fetch, and Reason: A Unified Evaluation of Retrieval-Augmented Generation

【417】 PoisonedRAG: Knowledge Corruption Attacks to Retrieval-Augmented Generation of Large Language Models

【418】 Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection

【419】 RAG-Anything: All-in-One RAG Framework

【420】 Chapter 5: Code as an Interaction Interface and Generating SQL Queries

【421】 README

【422】 agent.py

【423】 demo.py

【424】 aggregate functions

【425】 read-only URI mode

【426】 authorization callbacks

【427】 defenses for untrusted SQL

【428】 application-defined function security

【429】 query plans

【430】 type affinity

【431】 STRICT tables and version requirements

【432】 “SchemaGraphSQL: Efficient Schema Linking with Pathfinding Graph Algorithms for Text-to-SQL on Large-Scale Databases”

【433】 “Rethinking Schema Linking: A Context-Aware Bidirectional Retrieval Approach for Text-to-SQL”

【434】 agent.py

【435】 demo.py

【436】 demo.py comparison flow

【437】 OpenAI Agents SDK: Agents

【438】 LangChain: Short-term memory

【439】 MCP joins the Agentic AI Foundation

【440】 A2A Protocol: Core Concepts

【441】 MCP Governance and Stewardship

【442】 Microsoft Agent Framework Overview

【443】 AutoGen Maintenance Mode

【444】 LangChain: Context Engineering for Agents

【445】 Drew Breunig: How Contexts Fail and How to Fix Them

【446】 Anthropic: Code execution with MCP

【447】 Cloudflare: Code Mode — the better way to use MCP

【448】 RAG-MCP: Mitigating Prompt Bloat in LLM Tool Selection via Retrieval-Augmented Generation

【449】 AGENTS.md

【450】 OpenAI Codex: Custom instructions with AGENTS.md

【451】 Reflexion: Language Agents with Verbal Reinforcement Learning

【452】 Reflexion: original experimental setup and Tables 1–2

【453】 ReWOO: Decoupling Reasoning from Observations for Efficient Augmented Language Models

【454】 An LLM Compiler for Parallel Function Calling

【455】 LangChain: Planning Agents

【456】 v2, §4.1 and Table 2

【457】 Graph of Thoughts: Solving Elaborate Problems with Large Language Models

【458】 PAL: Program-aided Language Models

【459】 Self-Refine: Iterative Refinement with Self-Feedback

【460】 Plan-and-Solve Prompting

【461】 Program of Thoughts Prompting

【462】 Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting

【463】 Let's Verify Step by Step

【464】 Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters

【465】 Python 3.13: decimal — precision, rounding, and decimal arithmetic

【466】 AWS Builders' Library: Making retries safe with idempotent APIs

【467】 SHOP planners: the authors' project page and implementations

【468】 Martin Kleppmann: How to do distributed locking

【469】 CoALA: Cognitive Architectures for Language Agents

【470】 MemGPT: Towards LLMs as Operating Systems

【471】 Generative Agents: Interactive Simulacra of Human Behavior

【472】 LangGraph: Memory overview

【473】 documentation snapshot 1fa2214

【474】 OpenAI: Safety in building agents

【475】 LongMemEval paper

【476】 official README snapshot 9e0b455

【477】 LoCoMo paper

【478】 ACL 2024 release notes snapshot 9228632

【479】 LongMemEval-V2 official README snapshot 2cc8c54

【480】 LongMemEval

【481】 LoCoMo

【482】 LongMemEval-V2

【483】 LangGraph: Persistence

【484】 documentation snapshot 3edafc8

【485】 LangGraph: Checkpointers

【486】 AWS Prescriptive Guidance: Transactional outbox pattern

【487】 etcd v3.5: Transaction comparisons and revisions

【488】 Anthropic: How we built our multi-agent research system

【489】 Cognition: Don't Build Multi-Agents

【490】 Why Do Multi-Agent LLM Systems Fail? (MAST)

【491】 Raft: Algorithm and papers maintained by the authors

【492】 AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation

【493】 CAMEL: Communicative Agents for Mind Exploration of Large Language Model Society

【494】 Anthropic Prompt Caching

【495】 OpenAI Compaction

【496】 LangGraph Memory

【497】 LLMs Can't Plan, But Can Help Planning in LLM-Modulo Frameworks

【498】 Do As I Can, Not As I Say: Grounding Language in Robotic Affordances

【499】 OR-Tools: CP-SAT Solver

【500】 Reflexion: Language Agents with Verbal Reinforcement Learning (v4)

【501】 Language Agent Tree Search Unifies Reasoning, Acting, and Planning (v3)

【502】 Improving Factuality and Reasoning in Language Models through Multiagent Debate

【503】 Large Language Models Cannot Self-Correct Reasoning Yet

【504】 Training Language Models to Self-Correct via Reinforcement Learning (SCoRe, v1)

【505】 SCoRe

【506】 LangGraph Graph API

【507】 LangGraph: Concurrent State Update Error

【508】 OpenAI Agents SDK

【509】 OpenAI Swarm: Official Guidance on Migrating to the Agents SDK

【510】 OpenAI Agents SDK: Handoffs

【511】 Amazon SQS: At-Least-Once Delivery

【512】 etcd v3.6: API Consistency and Lease Guarantees

【513】 etcd v3.6: Lock and Transactional Protection

【514】 LangGraph: add_messages Implementation

【515】 Diego Ongaro's Thesis: Replicated State Machines and Failure Assumptions

【516】 SWE-Lancer: Can Frontier LLMs Earn $1 Million from Real-World Freelance Software Engineering?

【517】 Terminal-Bench: Official tasks and versions

【518】 GAIA: a benchmark for General AI Assistants

【519】 tau^2-Bench: Evaluating Conversational Agents in a Dual-Control Environment

【520】 BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents

【521】 AgentBench: Evaluating LLMs as Agents

【522】 MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering

【523】 AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents

【524】 OpenTelemetry: Generative AI Semantic Conventions

【525】 OpenTelemetry: GenAI convention status (Development, pinned commit 0c87594)

【526】 Anthropic: Demystifying evals for AI agents

【527】 SWE-bench: Evaluation harness

【528】 OpenAI: BrowseComp reference implementation

【529】 official evaluators

【530】 Simon Willison: The lethal trifecta for AI agents

【531】 Simon Willison: The Dual LLM pattern for building AI assistants that can resist prompt injection

【532】 Design Patterns for Securing LLM Agents against Prompt Injections (v1)

【533】 Defeating Prompt Injections by Design (CaMeL, v2)

【534】 The Task Shield: Enforcing Task Alignment to Defend Against Indirect Prompt Injection in LLM Agents

【535】 System-Level Defense against Indirect Prompt Injection Attacks: An Information Flow Control Perspective

【536】 Bypassing LLM Guardrails: An Empirical Analysis of Evasion Attacks against Prompt Injection and Jailbreak Detection Systems (v3)

【537】 Docker Engine: seccomp configuration and boundaries for Linux containers

【538】 SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering

【539】 METR: Measuring AI Ability to Complete Long Software Tasks

【540】 Simon Willison: Designing agentic loops

【541】 Claude Agent SDK: Overview

【542】 GitHub Docs: Configure the development environment for Copilot cloud agent

【543】 Graph API

【544】 Functional API

【545】 Claude Agent SDK: How the agent loop works

【546】 OpenAI Agents SDK: Running agents

【547】 LangGraph: Subgraphs

【548】 Claude Agent SDK: Modifying system prompts

【549】 OpenAI Agents SDK: Guardrails

【550】 Claude Agent SDK: Configure permissions

【551】 gVisor documentation

【552】 GitHub Docs: Customizing or disabling the firewall

【553】 LangGraph: Interrupts

【554】 LangGraph: Fault tolerance

【555】 Temporal: Retry Policies

【556】 Stripe: Idempotent requests

【557】 OpenTelemetry: Traces

【558】 OpenAI Codex repository

【559】 relevant Chapter 5 passages

【560】 edit_tool.py at the same commit

【561】 Language Server Protocol 3.17

【562】 Python ast.parse documentation

【563】 failure attribution, exact-copy diagnosis, and trajectory-prefix regression

【564】 multi-turn credit assignment, rewards, and on-policy distillation

【565】 from failure cases to post-training

【566】 README for the premature-completion DPO repair

【567】 LangChain documentation

【568】 LangGraph documentation

【569】 LlamaIndex documentation

【570】 CrewAI documentation

【571】 RunnableSequence: LCEL composition, batching, and streaming semantics

【572】 Runnable API

【573】 langchain-core Runnables API reference

【574】 LangChain v1 migration guide

【575】 LangChain: Agents concepts

【576】 LangChain: Messages concepts

【577】 LangChain: Tools concepts

【578】 LangChain: Middleware concepts

【579】 LangChain: Structured Output

【580】 LangChain: Long-term Memory

【581】 LangChain: Streaming

【582】 LangChain: Human-in-the-Loop Approval and Resumption

【583】 LangGraph Graph API: Recursion Limit

【584】 langchain-core Tools API Reference

【585】 Pydantic Documentation

【586】 LangGraph: Stores

【587】 LangChain: Retrieval

【588】 LlamaIndex: Building an Agent

【589】 LlamaIndex: Workflows

【590】 LangChain4j official documentation

【591】 LangChain4j GitHub repository

【592】 LangChain4j: AI Services tutorial

【593】 LangChain4j: Tools tutorial

【594】 LangChain4j: Chat Memory tutorial

【595】 LangChain4j: RAG tutorial

【596】 LangChain4j: Guardrails and execution order

【597】 LangChain4j: AI Service Observability

【598】 Quarkus LangChain4j

【599】 Spring AI official documentation

【600】 LangSmith documentation

【601】 LangGraph Python package dependencies

【602】 LangGraph: Time Travel

【603】 LangChain: Event streaming

【604】 LangChain v1 release notes

【605】 LangChain v0.3 official release notes: Pydantic 2 migration

【606】 LangChain official blog

【607】 Pydantic migration guide

【608】 LangGraph Graph API: graph migrations

【609】 LangChain official blog: Open Deep Research

【610】 Official open_deep_research repository

【611】 Deep Agents overview

【612】 Deep Agents Backends

【613】 Deep Agents Sandboxes

【614】 Deep Agents synchronous subagents

【615】 Deep Agents async subagents

【616】 Official deepagents repository

【617】 LangSmith Evaluation documentation

【618】 LangSmith dataset management

【619】 LangSmith Online evaluation

【620】 LangSmith user feedback

【621】 LangSmith: hide and transform sensitive data before transmission

【622】 LangSmith experiment repetitions, concurrency, and caching

【623】 LlamaIndex official documentation

【624】 LlamaIndex: Loading Data (Ingestion Pipeline)

【625】 LlamaIndex: Indexing concepts

【626】 LlamaIndex: Default and optional retrieval modes for each index

【627】 LlamaIndex: Property Graph Index

【628】 LlamaIndex: Storage concepts

【629】 from_params() does not discover an existing application connection

【630】 LlamaIndex: Query Engine concepts

【631】 LlamaIndex: Routers and selectors

【632】 LlamaIndex: Workflows

【633】 LlamaIndex: Shared state in Workflows

【634】 LlamaIndex: Durable Workflows

【635】 LlamaIndex: WorkflowServer deployment

【636】 LlamaIndex: BaseSynthesizer query / nodes interface

【637】 Official DSPy documentation

【638】 DSPy: Class-based signatures

【639】 DSPy: Changing modules

【640】 DSPy paper: Khattab et al., "DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines"

【641】 DSPy: Choosing an optimizer

【642】 DSPy: MIPROv2 API and its three-stage mechanism

【643】 DSPy: GEPA API and feedback contract

【644】 DSPy: GEPA optimization tutorial

【645】 GEPA paper: Agrawal et al., "GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning"

【646】 Official LangSmith documentation

【647】 Semantic Kernel official documentation: Introduction

【648】 Semantic Kernel: Kernel concepts

【649】 Semantic Kernel: Plugin concepts

【650】 Semantic Kernel: Planning concepts

【651】 Semantic Kernel: Filter concepts

【652】 Semantic Kernel official repository and successor-framework announcement, pinned commit

【653】 Semantic Kernel: Process Framework

【654】 Semantic Kernel: Agent Framework overview

【655】 Semantic Kernel: Agent Orchestration

【656】 Semantic Kernel: Complete Group Chat runtime example

【657】 SK → MAF migration guide

【658】 Microsoft: Semantic Kernel and MAF support transition announcement

【659】 AutoGen official documentation

【660】 AutoGen: Core user guide

【661】 AutoGen: AgentChat user guide

【662】 AutoGen official repository: Maintenance Mode, pinned commit

【663】 AutoGen → MAF migration guide

【664】 AutoGen: Saving and loading Agent/Team state

【665】 CrewAI: Flow concepts

【666】 CrewAI: Crew concepts

【667】 CrewAI: Sequential and hierarchical Processes

【668】 PydanticAI official documentation

【669】 PydanticAI: Core Agent concepts

【670】 PydanticAI: Function Tools

【671】 PydanticAI: Dependency injection

【672】 PydanticAI: Output modes and validation

【673】 PydanticAI: Durable Execution

【674】 OpenTelemetry Generative AI semantic conventions and status

【675】 Martin Fowler: StranglerFigApplication

【676】 Martin Fowler: CanaryRelease

【677】 Alistair Cockburn: Hexagonal Architecture

【678】 Google Cloud: MLOps: Continuous delivery and automation pipelines in machine learning

【679】 a16z: What Is LLMOps?

【680】 Chip Huyen: Building LLM applications for production

【681】 OpenAI: Best practices for production deployments

【682】 Google SRE Book: Chapter 1 - Introduction

【683】 Uber Engineering: Michelangelo Machine Learning Platform

【684】 Martin Fowler: Continuous Delivery for Machine Learning

【685】 LiteLLM: Routing

【686】 Amazon Bedrock: Model routing (intelligent prompt routing)

【687】 Netflix Tech Blog: Fault Tolerance in a High Volume, Distributed System

【688】 Google Cloud: Implementing exponential backoff

【689】 Martin Fowler: CircuitBreaker

【690】 AWS Well-Architected Framework: REL05-BP04 Bulkhead architecture

【691】 OpenAI: Structured Outputs

【692】 Anthropic: Tool use with Claude

【693】 JSON Schema Specification

【694】 Pydantic: Validators

【695】 OpenAI: Moderation API

【696】 Anthropic: Guardrails against misuse

【697】 Google Cloud: Responsible AI practices

【698】 OpenAI Evals

【699】 OpenAI: Evaluation best practices

【700】 OpenAI: 2026-06-03 Evals platform deprecation

【701】 SciPy: Binomial proportion confidence intervals

【702】 Google: Rules of Machine Learning - Rule #4: Keep the first model simple and get the infrastructure right

【703】 Braintrust: What is an eval?

【704】 LangSmith: Evaluation concepts

【705】 OpenTelemetry: Semantic conventions for generative AI systems

【706】 Google SRE Book: Monitoring Distributed Systems

【707】 Arize Phoenix: Tracing

【708】 Honeycomb: Observability for LLMs

【709】 OpenAI: GPT-4o snapshots

【710】 OpenAI: Deprecations

【711】 DVC: Data Version Control

【712】 MLflow: Model Registry

【713】 Google SRE Workbook: Canarying Releases

【714】 SciPy: ttest_ind, independent samples, and Welch’s test assumptions

【715】 Spinnaker: Canary Analysis

【716】 Optimizely: Statistical significance in A/B testing

【717】 OpenAI: Prompt caching

【718】 OpenAI: Batch API

【719】 Anthropic: Message Batches API

【720】 Google Cloud: Cost optimization for AI and ML workloads

【721】 Google SRE Book: Service Level Objectives

【722】 Google SRE Workbook: Implementing SLOs

【723】 OpenAI: Rate limits

【724】 Anthropic: Rate limits

【725】 PagerDuty: Incident Response Documentation

【726】 OpenAI: Fine-tuning

【727】 Anthropic: Constitutional AI and RLHF

【728】 Netflix Tech Blog: Recommendation systems and the data flywheel

【729】 Google: People + AI Guidebook - Feedback + Control

【730】 MITRE ATLAS

【731】 NIST Generative AI Profile (NIST AI 600-1)

【732】 OWASP Agentic AI Threats and Mitigations

【733】 OWASP Top 10 for Agentic Applications 2026

【734】 NIST AI 100-2 E2025: Adversarial Machine Learning Taxonomy

【735】 OWASP LLM01:2025 Prompt Injection

【736】 OWASP LLM07:2025 System Prompt Leakage

【737】 Many-shot Jailbreaking (Anthropic)

【738】 Universal and Transferable Adversarial Attacks on Aligned Language Models

【739】 Ignore This Title and HackAPrompt: Exposing Systemic Vulnerabilities of LLMs

【740】 MITRE ATLAS: Prompt Injection

【741】 OWASP LLM05:2025 Improper Output Handling

【742】 OWASP LLM02:2025 Sensitive Information Disclosure

【743】 Imprompter: Tricking LLM Agents into Improper Tool Use

【744】 OWASP: CSV Injection

【745】 Poisoning Web-Scale Training Datasets is Practical

【746】 Anthropic / UK AISI / Alan Turing Institute: Small samples can poison language models

【747】 BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chain

【748】 A Backdoor Attack Against LSTM-based Text Classification Systems

【749】 On the Exploitability of Instruction Tuning

【750】 OWASP LLM04:2025 Data and Model Poisoning

【751】 OWASP LLM03:2025 Supply Chain

【752】 Hugging Face: Pickle Scanning and Safetensors

【753】 PyTorch 2.6: Serialization semantics and weights_only

【754】 Sleepy Pickle: Exploiting Machine Learning Pickle Files

【755】 MITRE ATLAS: ML Supply Chain Compromise

【756】 CycloneDX: Machine Learning Bill of Materials (ML-BOM)

【757】 Sigstore: Software Signing for Everyone

【758】 Extracting Training Data from Large Language Models

【759】 Quantifying Memorization Across Neural Language Models

【760】 Membership Inference Attacks against Machine Learning Models

【761】 Deep Learning with Differential Privacy (DP-SGD)

【762】 GDPR original text: Articles 5, 6, 17, and Chapter V

【763】 Confused Deputy Problem (Norm Hardy, 1988)

【764】 RFC 8693: OAuth 2.0 Token Exchange, especially Sections 1.1 and 4.1

【765】 RFC 7662: OAuth 2.0 Token Introspection

【766】 NIST SP 800-207: Zero Trust Architecture

【767】 SPIFFE/SPIRE: Workload Identity Framework

【768】 OWASP LLM06:2025 Excessive Agency

【769】 gVisor: Application Kernel for Containers

【770】 Firecracker: Secure and Fast microVMs

【771】 MDN: Same-origin policy

【772】 Anthropic: Computer Use Demo — security precautions and isolation limitations

【773】 garak: LLM Vulnerability Scanner

【774】 PyRIT: Python Risk Identification Tool for generative AI

【775】 HarmBench: A Standardized Evaluation Framework for Automated Red Teaming

【776】 Red Teaming Language Models with Language Models

【777】 OWASP LLM Applications Cybersecurity and Governance Checklist

【778】 EU AI Act: original text of Regulation (EU) 2024/1689

【779】 Regulation (EU) 2026/1744: Article 1, points (39) and (40), and Article 4

【780】 European Commission: AI Act policy overview

【781】 C2PA 2.2: Explainer, verification capabilities and non-goals

【782】 Model Cards for Model Reporting

【783】 System Cards: A New Resource for Understanding How AI Systems Work

【784】 Palantir: Dev versus Delta—Demystifying Engineering Roles at Palantir

【785】 Palantir: A Day in the Life of a Forward Deployed Software Engineer

【786】 Palantir: AIP Chatbot Studio Overview

【787】 Palantir: Security and Governance

【788】 OpenAI Deployment Company: Build, Prove, Generalize

【789】 OpenAI: Forward Deployed Engineer, Gov

【790】 OpenAI: Production Best Practices

【791】 Anthropic: How We Built Claude Managed Agents

【792】 Palantir: Securing Software at the Speed of AI—Official Engineering Case, 2026

【793】 Microsoft: Only Believe What You Can Validate—Customer Field Engineering Account, 2026

【794】 Baseten: Forward Deployed Engineering on the Frontier of AI—Official Team Practices, 2025

【795】 AWS and INRIX: Transportation-Planning PoC—Coauthored by Customer and Vendor, 2025

【796】 @vasuman: Forward Deployed Engineering 101—Practitioner's Long-Form Article on X

【797】 Hamel Husain: A Field Guide to Rapidly Improving AI Products

【798】 Varick Agents: Careers

【799】 run_maotui: Working with AI in Product Management

【800】 Applied AI Field Guide

【801】 Five-minute introduction

【802】 FDEstack

【803】 Overview of eight skills

【804】 OpenFDE

【805】 Architecture

【806】 FDEOps

【807】 Project overview

【808】 invoice-exception teaching case

【809】 hold-scope

【810】 unknowns.md

【811】 /poc

【812】 /integrate

【813】 handoff process

【814】 eval command

【815】 task transitions

【816】 handoff case

【817】 pattern capture

【818】 Claude extraction implementation

【819】 privacy notes

Chapter reading notes

Chapter 1: How Large Language Models Differ from Traditional NLP

  • BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding【1】
  • Language Models are Few-Shot Learners (GPT-3)【2】
  • Language Models are Unsupervised Multitask Learners (GPT-2)【3】
  • Emergent Abilities of Large Language Models【4】
  • Are Emergent Abilities of Large Language Models a Mirage?【5】
  • Scaling Laws for Neural Language Models【6】
  • Training Compute-Optimal Large Language Models (Chinchilla)【7】
  • The Llama 3 Herd of Models【8】

Chapter 2: Transformer Architecture

  • Attention Is All You Need【9】
  • BERT【1】
  • Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer (T5)【10】
  • On Layer Normalization in the Transformer Architecture【11】
  • Root Mean Square Layer Normalization【12】
  • GLU Variants Improve Transformer【13】
  • Query-Key Normalization for Transformers【14】
  • Transformer Feed-Forward Layers Are Key-Value Memories【15】

Chapter 3: MHA Limitations, MQA, GQA, and FlashAttention

  • Fast Transformer Decoding: One Write-Head is All You Need (MQA)【16】
  • GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints【17】
  • Llama 2【18】
  • The Llama 3 Herd of Models【8】
  • DeepSeek-V2 paper and official implementation【19】
  • FlashAttention【20】
  • FlashAttention official repository and version requirements (README snapshot, 2026-07-06)【21】
  • Online normalizer calculation for softmax【22】
  • Mistral 7B【23】
  • Linformer【24】
  • Rethinking Attention with Performers【25】
  • Mamba: Linear-Time Sequence Modeling with Selective State Spaces【26】

Chapter 4: Positional Encoding

  • Attention Is All You Need【9】
  • RoFormer: Enhanced Transformer with Rotary Position Embedding【27】
  • ALiBi: Train Short, Test Long【28】
  • Position Interpolation【29】
  • YaRN: Efficient Context Window Extension of Large Language Models【30】
  • Hugging Face Transformers: RoPE parameters and variants【31】
  • BLOOM model card【32】
  • MosaicML: MPT-7B release announcement and ALiBi configuration【33】
  • Lost in the Middle: How Language Models Use Long Contexts【34】

Chapter 5: Tokenizers

  • Neural Machine Translation of Rare Words with Subword Units (BPE)【35】
  • SentencePiece: A simple and language independent subword tokenizer【36】
  • Subword Regularization: Improving NMT Models with Multiple Subword Candidates (Unigram)【37】
  • Hugging Face: Tokenizers tutorial【38】
  • OpenAI tiktoken【39】
  • The Llama 3 Herd of Models【8】
  • Meta Llama 3 tokenizer implementation【40】
  • Google SentencePiece implementation and configuration【41】
  • Hugging Face Transformers: Tokenization algorithms【42】
  • Hugging Face Transformers: Chat templates【43】

Chapter 6: The Three-Stage Training Framework for LLMs

  • GPT-3, §2 and few-shot evaluations【44】
  • Llama 2, §§2.1 and 3.1.1, and the model card【45】
  • The Llama 3 Herd of Models, §§3–4【46】
  • Chinchilla: the training-compute approximation【47】
  • InstructGPT【48】
  • LIMA【49】
  • Deep Reinforcement Learning from Human Preferences【50】
  • DPO: reward reparameterization and preference loss【51】
  • DeepSeek-R1: the training workflow in the initial 2025 report【52】

Chapter 7: Scaling Laws and Emergent Abilities

  • Scaling Laws for Neural Language Models【6】
  • Chinchilla: §3, Table 2, and the compute-optimal derivation【47】
  • Emergent Abilities of Large Language Models【4】
  • Are Emergent Abilities of Large Language Models a Mirage?【5】
  • The Llama 3 Herd of Models【46】
  • Qwen3 Technical Report, initial version【53】
  • DeepSeekMath: data construction and GRPO【54】
  • DeepSeek-R1, initial version【52】

Chapter 8: Fine-Tuning Approaches for LLMs

Chapter 9: LoRA in Depth

  • LoRA: equations, initialization, and rank experiments【55】
  • QLoRA【56】
  • Parameter-Efficient Transfer Learning for NLP【59】
  • PEFT v0.17.0: LoRA configuration, initialization, and scaling【60】
  • PEFT: Model merging and add_weighted_adapter【61】
  • PEFT v0.17.0: weighted-adapter implementation and limitations【62】
  • vLLM: LoRA adapters【63】
  • Transformers: memory components of model training【57】

Chapter 10: A Survey of Post-Training Methods

Chapter 11: DPO vs. PPO in Depth

  • Proximal Policy Optimization Algorithms【66】
  • OpenAI Spinning Up: PPO-Clip equations and constraint limits【67】
  • InstructGPT【48】
  • DPO: §4 and Appendix A derivations【51】
  • DeepSeekMath: GRPO【54】
  • DeepSeek-R1, initial version【52】
  • Llama 2【45】
  • The Llama 3 Herd of Models【46】
  • High-Dimensional Continuous Control Using Generalized Advantage Estimation【68】

Chapter 12: Decoding Strategies

  • The Curious Case of Neural Text Degeneration【69】
  • If Beam Search Is the Answer, What Was the Question?【70】
  • Fast Inference from Transformers via Speculative Decoding【71】
  • Self-Consistency Improves Chain of Thought Reasoning in Language Models【72】
  • PagedAttention: cache sharing, including beam search【73】
  • Qwen3-30B-A3B official model card【74】
  • vLLM: Batch Invariance【75】

Chapter 13: Tuning Temperature, Top-P, and Top-K

  • The Curious Case of Neural Text Degeneration【69】
  • Transformers: Generation Utilities, including Temperature/TopK/TopP LogitsWarper【76】
  • Transformers v4.56.2: configuration conditions for greedy decoding, sampling, and beam search【77】
  • Qwen3-30B-A3B official model card and sampling recommendations【74】
  • vLLM: Batch Invariance【75】

Chapter 14: KV Cache and Prompt Caching

Interface review scope in the source manuscript: official documentation and SDK as of 2026-09-08, with the OpenAI caching guide checked again on 2026-09-15. The example did not make a billable request and does not report a real cache hit.

Chapter 15: Model Quantization

  • GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers【83】
  • AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration【84】
  • QLoRA: Efficient Finetuning of Quantized LLMs (NF4)【56】
  • LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale【85】
  • SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models【86】
  • Optimal Brain Compression: A Framework for Accurate Post-Training Quantization and Pruning【87】
  • llama.cpp / GGUF【88】
  • AutoAWQ【89】
  • GPTQ authors' implementation: column order, act-order, and grouping【90】
  • AWQ paper: per-channel scaling and search【91】
  • PyTorch AO: Quantization-Aware Training【92】
  • FP8 Formats for Deep Learning, v1, §§2–3 and Table 1【93】
  • OCP Microscaling Formats (MX) Specification v1.0【94】
  • Microsoft MX library: max-exponent scaling and scale limits, pinned revision【95】
  • NVIDIA: Introducing NVFP4 for Efficient and Accurate Low-Precision Inference【96】
  • NVIDIA Transformer Engine: FP8 primer and MXFP8/NVFP4, pinned notebook【97】
  • NVIDIA Transformer Engine: scaling recipes, history, and margin, pinned source【98】
  • gpt-oss-120b & gpt-oss-20b model card, v1, §2.1 and Table 1: MXFP4 MoE weights and checkpoint sizes【99】
  • AMD ROCm blog: FP8 FNUZ and OCP variants on CDNA3 and CDNA4【100】
  • ONNX: FP8 normal/subnormal decoding and special-value encodings【101】
  • vLLM v0.10.2: E4M3FN-to-E4M3FNUZ normalization, pinned implementation【102】

Chapter 16: Prompt Engineering

  • Language Models are Few-Shot Learners (GPT-3, few-shot learning)【2】
  • Calibrate Before Use: Improving Few-Shot Performance of Language Models (example and ordering biases)【103】
  • Chain-of-Thought Prompting Elicits Reasoning in Large Language Models【104】
  • Large Language Models are Zero-Shot Reasoners (Let's think step by step)【105】
  • LLMLingua: Compressing Prompts for Accelerated Inference of Large Language Models【106】
  • LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compression【107】
  • The Power of Scale for Parameter-Efficient Prompt Tuning (soft prompts)【108】
  • OpenAI: Reasoning best practices (including the scope of delimiters and CoT prompts)【109】
  • OpenAI: Structured Outputs【110】

Chapter 17: Chain of Thought

  • Chain-of-Thought Prompting Elicits Reasoning in Large Language Models【104】
  • Large Language Models are Zero-Shot Reasoners (Let's think step by step)【105】
  • Self-Consistency Improves Chain of Thought Reasoning in Language Models【72】
  • Least-to-Most Prompting Enables Complex Reasoning in Large Language Models【111】
  • Tree of Thoughts: Deliberate Problem Solving with Large Language Models【112】
  • Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matters【113】
  • Measuring Faithfulness in Chain-of-Thought Reasoning【114】
  • s1: Simple test-time scaling【115】
  • Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs【116】
  • OpenAI: Reasoning best practices【109】
  • OpenAI: Reasoning models【117】
  • Anthropic Python SDK: manual thinking configuration【118】
  • Amazon Bedrock: Extended thinking【119】 (covers Bedrock's manual-budget rules, the interleaved-thinking exception, and model differences; it does not imply identical behavior across every Claude access path)

Chapter 18: Hallucinations—Causes and Mitigation

  • On Faithfulness and Factuality in Abstractive Summarization【120】
  • Survey of Hallucination in Natural Language Generation【121】
  • A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions【122】
  • Language Models (Mostly) Know What They Know【123】
  • TruthfulQA: Measuring How Models Mimic Human Falsehoods【124】
  • Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks【125】
  • SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models【126】
  • Chain-of-Verification Reduces Hallucination in Large Language Models【127】
  • Constitutional AI: Harmlessness from AI Feedback【64】
  • Towards Understanding Sycophancy in Language Models【128】
  • OpenAI: Structured Outputs (structural guarantees versus content errors)【110】

Chapter 19: Mixture-of-Experts Models

Chapter 20: Choosing a Serving Framework

  • Efficient Memory Management for Large Language Model Serving with PagedAttention (vLLM)【73】
  • SGLang: Efficient Execution of Structured Language Model Programs (RadixAttention)【137】
  • Orca: A Distributed Serving System for Transformer-Based Generative Models (continuous batching)【138】
  • vLLM documentation【139】
  • vLLM: Automatic Prefix Caching【140】
  • SGLang repository【141】
  • Text Generation Inference repository【142】
  • TGI official documentation: maintenance-mode notice, checked in the source manuscript on 2026-09-15【143】
  • llama.cpp repository【88】
  • TensorRT-LLM repository【144】
  • llama.cpp: HTTP server capabilities and parameters【145】
  • TensorRT-LLM: Quick Start【146】
  • vLLM: Chunked Prefill and configuration tuning【147】
  • vLLM: Disaggregated Prefilling【148】
  • TGI: components and HTTP/gRPC boundaries【149】
  • TensorRT-LLM: LLM API and PyTorch backend【150】

Chapter 21: Capability Evaluation Metrics

  • Measuring Massive Multitask Language Understanding (MMLU)【151】
  • MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark【152】
  • Evaluating Large Language Models Trained on Code (HumanEval / Pass@k)【153】
  • SWE-bench: Can Language Models Resolve Real-World GitHub Issues?【154】
  • OpenAI: Why SWE-bench Verified no longer measures frontier coding capabilities【155】
  • Training Verifiers to Solve Math Word Problems (GSM8K)【156】
  • Measuring Mathematical Problem Solving With the MATH Dataset【157】
  • GPQA: A Graduate-Level Google-Proof Q&A Benchmark【158】
  • Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena【159】
  • Holistic Evaluation of Language Models (HELM)【160】
  • LiveBench: A Challenging, Contamination-Limited LLM Benchmark【161】
  • tau-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains【162】
  • HumanEval's official Pass@k implementation【163】
  • Official SWE-bench documentation: Verified subset【164】
  • Humanity's Last Exam【165】
  • On Faithfulness and Factuality in Abstractive Summarization (lexical metrics versus faithfulness)【120】

Chapter 22: Practical Model Selection

  • Holistic Evaluation of Language Models (HELM)【160】
  • Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena【159】
  • DeepSeek-V3 Technical Report【166】
  • Qwen3 Technical Report【167】
  • tau-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains【162】
  • RouteLLM: Learning to Route LLMs with Preference Data【168】
  • FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance【169】
  • OpenAI: Data controls in the OpenAI platform【170】
  • OpenAI: Structured Outputs【110】

Chapter 23: Multimodal Models

  • CLIP: Learning Transferable Visual Models From Natural Language Supervision【171】
  • Flamingo: a Visual Language Model for Few-Shot Learning【172】
  • LLaVA: Visual Instruction Tuning【173】
  • Whisper: Robust Speech Recognition via Large-Scale Weak Supervision【174】
  • Video-LLaMA: An Instruction-tuned Audio-Visual Language Model【175】
  • NIST AI Risk Management Framework【176】
  • BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models【177】
  • Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution【178】
  • Hugging Face Evaluate: WER definition and implementation notes【179】

Chapter 1: Multimodal Representations and Fusion Architectures

  • CLIP: Learning Transferable Visual Models From Natural Language Supervision【171】
  • Sigmoid Loss for Language Image Pre-Training (SigLIP)【180】
  • ImageBind: One Embedding Space To Bind Them All【181】
  • Flamingo: a Visual Language Model for Few-Shot Learning【172】
  • Perceiver IO: A General Architecture for Structured Inputs & Outputs【182】
  • LLaVA: Visual Instruction Tuning【173】
  • Qwen-VL: Position-aware vision–language adapter【183】
  • Chameleon: Mixed-Modal Early-Fusion Foundation Models【184】
  • Neural Discrete Representation Learning (VQ-VAE)【185】
  • Taming Transformers for High-Resolution Image Synthesis (VQGAN)【186】
  • OpenAI GPT-4o System Card【187】
  • GPT-4o System Card (original report)【188】

Chapter 2: Vision–Language Models and Visual Grounding

  • Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection【189】
  • Grounded Language-Image Pre-training (GLIP)【190】
  • Simple Open-Vocabulary Object Detection with Vision Transformers (OWL-ViT)【191】
  • PaliGemma: A versatile 3B VLM for transfer【192】
  • Qwen-VL: A Versatile Vision-Language Model【183】
  • Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Multimodal Models【193】
  • Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations【194】
  • Generation and Comprehension of Unambiguous Object Descriptions (RefCOCOg)【195】
  • ReferItGame: Referring to Objects in Photographs of Natural Scenes【196】
  • REFER: Dataset origins and split documentation【197】

Chapter 3: OCR and Document AI

  • LayoutLM: Pre-training of Text and Layout for Document Image Understanding【198】
  • LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking【199】
  • OCR-free Document Understanding Transformer (Donut)【200】
  • PubTables-1M: Towards Comprehensive Table Extraction From Unstructured Documents【201】
  • GriTS: Grid Table Similarity【202】
  • Image-based table recognition: data, model, and evaluation (PubTabNet / TEDS)【203】
  • ChartQA: A Benchmark for Question Answering about Charts【204】
  • DocVQA: A Dataset for VQA on Document Images【205】
  • Azure AI Document Intelligence documentation【206】
  • Azure Document Intelligence: Accuracy and confidence【207】
  • Google Cloud Document AI documentation【208】
  • Amazon Textract documentation【209】
  • Amazon Textract: AnalyzeDocument and FeatureTypes【210】

Chapter 4: Computer Use and GUI Agents

  • Anthropic: Developing a computer use model【211】
  • Anthropic: Upgraded Claude 3.5 Sonnet and computer use public beta【212】
  • Anthropic Computer Use documentation【213】
  • OpenAI: Computer-Using Agent【214】
  • OpenAI Operator【215】
  • OpenAI: Computer use tools and execution environments【216】
  • UI-TARS: Pioneering Automated GUI Interaction with Native Agents【217】
  • OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments【218】
  • WebArena: A Realistic Web Environment for Building Autonomous Agents【219】

Chapter 5: Speech Recognition, Synthesis, and Audio Language Models

  • Whisper: Robust Speech Recognition via Large-Scale Weak Supervision【174】
  • Connectionist Temporal Classification (original paper)【220】
  • Sequence Transduction with Recurrent Neural Networks (RNN-T)【221】
  • Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions (Tacotron 2)【222】
  • Neural Codec Language Models are Zero-Shot Text to Speech Synthesizers (VALL-E)【223】
  • SoundStream: An End-to-End Neural Audio Codec【224】
  • High Fidelity Neural Audio Compression (EnCodec)【225】
  • AudioLM: a Language Modeling Approach to Audio Generation【226】
  • Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models【227】
  • SALMONN: Towards Generic Hearing Abilities for Large Language Models【228】

Chapter 6: Real-Time Full-Duplex Voice Interaction

  • Moshi: a speech-text foundation model for real-time dialogue【229】
  • Generative Spoken Dialogue Language Modeling (dGSLM)【230】
  • OpenAI GPT-4o System Card【187】
  • GPT-4o System Card (original report)【188】
  • OpenAI Realtime API documentation【231】
  • OpenAI Realtime: Conversation state and interruption handling【232】
  • Kyutai Moshi project page【233】

Chapter 7: Diffusion Models, Flow Matching, and Image Generation

  • Denoising Diffusion Probabilistic Models (DDPM)【234】
  • Denoising Diffusion Implicit Models (DDIM)【235】
  • Score-Based Generative Modeling through Stochastic Differential Equations【236】
  • High-Resolution Image Synthesis with Latent Diffusion Models (Stable Diffusion)【237】
  • Classifier-Free Diffusion Guidance【238】
  • Flow Matching for Generative Modeling【239】
  • Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow【240】
  • Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (Stable Diffusion 3)【241】
  • Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding (Imagen)【242】
  • OpenAI DALL·E 3: Improving Image Generation with Better Captions【243】
  • GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium (FID)【244】
  • CLIPScore: A Reference-free Evaluation Metric for Image Captioning【245】

Chapter 8: Video Generation Models

  • Video Diffusion Models【246】
  • Imagen Video: High Definition Video Generation with Diffusion Models【247】
  • OpenAI: Video generation models as world simulators (Sora technical report)【248】
  • Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets【249】
  • Towards Accurate Generative Models of Video: A New Metric & Challenges (FVD)【250】
  • VBench: Comprehensive Benchmark Suite for Video Generative Models【251】

Chapter 9: Multimodal Training Data and Alignment

  • LAION-5B: An Open Large-Scale Dataset for Training Next Generation Image-Text Models【252】
  • OpenAI DALL·E 3: Improving Image Generation with Better Captions【243】
  • LLaVA: Visual Instruction Tuning【173】
  • Flamingo: a Visual Language Model for Few-Shot Learning【172】
  • OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents【253】
  • Aligning Large Multimodal Models with Factually Augmented RLHF (LLaVA-RLHF)【254】

Chapter 10: Multimodal Evaluation, Safety, and Inference Serving

  • MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark【255】
  • MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts【256】
  • Multimodal Neurons in Artificial Neural Networks (Typographic Attack)【257】
  • Visual Adversarial Examples Jailbreak Aligned Large Language Models【258】
  • C2PA 2.1 technical specification【259】
  • Google DeepMind: SynthID【260】
  • DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving【261】
  • vLLM: Prefix caching design and multimodal hashing【82】

Chapter 1: What Function Calling Is and How It Works

  • OpenAI: Function Calling guide (checked 2026-09-08)【262】
  • OpenAI: Responses API migration guide (this section's fields and examples were checked against this page and the Function Calling guide on 2026-09-16)【263】
  • OpenAI: The 2023 Function Calling announcement【264】
  • Anthropic: Tool Use with Claude【265】
  • Anthropic: Writing Effective Tools for Agents【266】
  • Toolformer: Language Models Can Teach Themselves to Use Tools【267】
  • ReAct: Synergizing Reasoning and Acting in Language Models【268】
  • JSON Schema specification【269】

Chapter 2: How LLMs Learn to Use Tools

  • Toolformer: Language Models Can Teach Themselves to Use Tools【267】
  • ReAct paper【268】
  • PPO paper【66】
  • ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs【270】
  • Training language models to follow instructions with human feedback (InstructGPT)【48】
  • Constitutional AI: Harmlessness from AI Feedback【64】
  • ToolRL: Reward is All Tool Learning Needs【271】
  • ReTool: Reinforcement Learning for Strategic Tool Use in LLMs【272】
  • Berkeley Function Calling Leaderboard【273】
  • API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs【274】
  • τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains【162】

Chapter 3: Tool Definitions and Schema Engineering

  • Anthropic: Writing Effective Tools for Agents【266】
  • Anthropic: Building Effective Agents【275】
  • OpenAI: Function Calling guide【262】
  • OpenAI: Schemas supported by Structured Outputs【110】
  • JSON Schema specification【269】
  • Berkeley Function Calling Leaderboard【273】

Chapter 4: The Core of the Model Context Protocol

  • Version-status review: the original manuscript rechecked the following pinned-version changes and compatibility notes on 2026-09-15. It uses a fixed protocol baseline; a dynamic page saying “latest” is not evidence of SDK support.
  • MCP version status【276】
  • MCP 2026-07-28 changelog【277】
  • Model Context Protocol documentation【278】
  • MCP specification 2026-07-28【279】
  • MCP version compatibility【280】
  • MCP architecture【281】
  • Anthropic: Introducing the Model Context Protocol【282】
  • JSON-RPC 2.0 specification【283】

Chapter 5: The Three Layers of MCP

Chapter 6: MCP and Function Calling—Differences and Tradeoffs

  • Model Context Protocol documentation【278】
  • MCP Server development quickstart【290】
  • MCP Servers official examples【291】
  • Archived MCP examples【292】
  • MCP Roots: deprecation and the absence of a security boundary【293】
  • OpenAI: Function calling guide【294】
  • Anthropic: Introducing the Model Context Protocol【282】

Chapter 7: Why Some Reasoning Models Seem Not to Support MCP

  • OpenAI: Reasoning Models guide【117】
  • OpenAI: Function Calling and returning reasoning items【262】
  • OpenAI: o1-preview model page【295】
  • OpenAI: o1 model and snapshots【296】
  • OpenAI Cookbook: Early o1-preview limitations on structured outputs【297】
  • Anthropic: Claude 4 announcement of interleaved tool use【298】
  • Anthropic Cookbook: Early Claude 3.7 example of thinking-block preservation【299】
  • Anthropic: Extended Thinking【300】
  • DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning【301】
  • ReTool: Reinforcement Learning for Strategic Tool Use in LLMs【272】
  • ToolRL: Reward is All Tool Learning Needs【271】
  • Efficient Memory Management for Large Language Model Serving with PagedAttention【73】

Chapter 8: What Is a Skill?

  • The source manuscript retains commit 69ef37e9424c0a7ea9dd2293b559e43ec8176379 as its historical format baseline and records a 2026-09-15 review of required fields, experimental fields, and loading recommendations on the official format page. The official page has no separate semantic version number. This translation retains that pin; the pinned specification and current format page were checked again on 2026-09-20.
  • Anthropic: Introducing Agent Skills【302】
  • Anthropic: Equipping Agents for the Real World with Agent Skills【303】
  • Agent Skills specification【304】
  • Agent Skills pinned commit used for the format review【305】
  • Claude Docs: Agent Skills【306】
  • Anthropic: Building Effective Agents【275】
  • Anthropic: Effective Context Engineering for AI Agents【307】

Chapter 9: Skills and MCP

  • Anthropic: Introducing Agent Skills【302】
  • Agent Skills specification【304】
  • Model Context Protocol official documentation【278】
  • MCP 2026-07-28 tool discovery and schemas【285】
  • Anthropic: Equipping Agents for the Real World with Agent Skills【303】
  • Anthropic: Effective Context Engineering for AI Agents【307】

Chapter 10: How Function Calling, MCP, and Skills Fit Together

Chapter 11: The A2A Protocol

  • Source review: the Chinese manuscript's 2026-09-15 review read the v1.0.1 release notes, specification, and Proto definitions. The English migration rechecked the release metadata through GitHub's API and the relevant tagged specification and Proto sections. The release web page did not render through the retrieval tool. The tagged specification still contains an outdated “Latest Released Version 1.0.0” notice. Its Send Message overview also says the operation returns immediately, contradicting the execution-mode section and the Proto definition of blocking by default. This chapter follows the explicit return_immediately definition in the latter two sources; the JSON wire field is returnImmediately.
  • A2A v1.0.1 release notes【309】
  • A2A protocol website【310】
  • A2A v1.0.1 release specification【311】
  • A2A v1.0.1 protocol data definitions【312】
  • Google: Announcing the Agent2Agent Protocol【313】
  • Linux Foundation: A2A Project【314】
  • Model Context Protocol documentation【278】
  • Anthropic: Building Effective Agents【275】

Chapter 12: MCP Transports

Chapter 13: SSE, WebSocket, and WebRTC

Chapter 14: LLM Gateways

Chapter 15: Tool Protocol Security

  • MCP 2026-07-28 Authorization【342】
  • MCP 2026-07-28 Security Best Practices【343】
  • MCP 2026-07-28 Transports【288】
  • A2A v1.0.1 released specification【311】
  • RFC 9728: OAuth 2.0 Protected Resource Metadata【344】
  • RFC 8252: OAuth for Native Apps and loopback redirects【345】
  • RFC 9700: OAuth 2.0 Security Best Current Practice【346】
  • RFC 7636: PKCE【347】
  • RFC 8707: Resource Indicators for OAuth 2.0【348】

Chapter 1: What RAG Is and What Problems It Solves

  • Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks【125】
  • Dense Passage Retrieval for Open-Domain Question Answering【349】
  • Retrieval-Augmented Generation for Large Language Models: A Survey【350】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】
  • Anthropic: Introducing Contextual Retrieval【352】

Chapter 2: Tradeoffs Between RAG, Fine-Tuning, and Long Context

  • Lost in the Middle: How Language Models Use Long Contexts【34】
  • Chroma Research: Context Rot — How Increasing Input Tokens Impacts LLM Performance【353】
  • Can Long-Context Language Models Subsume Retrieval, RAG, SQL, and More?【354】
  • Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG【355】
  • LongRAG: Enhancing Retrieval-Augmented Generation with Long-context LLMs【356】
  • RAFT: Adapting Language Model to Domain Specific RAG【357】
  • Anthropic: Introducing Contextual Retrieval【352】

Chapter 3: Document Parsing and Preprocessing

  • ColPali: Efficient Document Retrieval with Vision Language Models【358】
  • Qdrant: Payload Updates and Indexing【359】
  • Advanced ingestion process powered by LLM parsing for RAG system【360】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】
  • Retrieval-Augmented Generation for Large Language Models: A Survey【350】

Chapter 4: Chunking Strategies and Granularity

  • Searching for Best Practices in Retrieval-Augmented Generation【351】
  • Dense X Retrieval: What Retrieval Granularity Should We Use?【361】
  • Qwen3-Embedding-0.6B Model Card【362】
  • Anthropic: Introducing Contextual Retrieval【352】
  • Retrieval-Augmented Generation for Large Language Models: A Survey【350】

Chapter 5: When Chunking Breaks the Meaning

  • Anthropic: Introducing Contextual Retrieval【352】
  • Late Chunking: Contextual Chunk Embeddings Using Long-Context Embedding Models【363】
  • Jina AI: Post-Encoding Pooling and Boundaries in Late Chunking【364】
  • Dense X Retrieval: What Retrieval Granularity Should We Use?【361】
  • RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval【365】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】

Chapter 6: Embedding Principles and Technical Evolution

  • Efficient Estimation of Word Representations in Vector Space【367】
  • Stanford GloVe: global co-occurrence statistics and a weighted least-squares objective【368】
  • BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding【1】
  • Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks【369】
  • SimCSE: Simple Contrastive Learning of Sentence Embeddings【370】
  • Dense Passage Retrieval for Open-Domain Question Answering【349】
  • ColBERTv2: Effective and Efficient Retrieval via Lightweight Late Interaction【371】
  • ColPali: Efficient Document Retrieval with Vision Language Models【358】

Chapter 7: Selecting and Evaluating Embedding Models

  • MTEB: Massive Text Embedding Benchmark【372】
  • MMTEB: Massive Multilingual Text Embedding Benchmark【373】
  • Matryoshka Representation Learning【374】
  • OpenAI: New embedding models and API updates (2024-01-25)【375】
  • OpenAI: Embedding API and the dimensions parameter【376】
  • Qwen3-Embedding-0.6B model card【362】
  • M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation【377】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】

Chapter 8: Vector Databases and ANN Indexes

  • Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs【378】
  • pgvector official README: filtering, iterative scans, VACUUM, and scaling【379】
  • FreshDiskANN: A Fast and Accurate Graph-Based ANN Index for Streaming Similarity Search【380】
  • ACORN: Performant and Predicate-Agnostic Search Over Vector Embeddings and Structured Data【381】
  • Billion-scale similarity search with GPUs【382】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】

Chapter 9: Vector Databases in Production and Performance Tuning

  • Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs【378】
  • hnswlib v0.8.0: internal IDs, base-layer adjacency capacity, and payload layout【383】
  • pgvector: exact baselines, filtering, and iterative scans【379】
  • FreshDiskANN: A Fast and Accurate Graph-Based ANN Index for Streaming Similarity Search【380】
  • Matryoshka Representation Learning【374】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】

Chapter 10: The Complete Online RAG Pipeline

  • Retrieval-Augmented Generation for Large Language Models: A Survey【350】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】
  • Lost in the Middle: How Language Models Use Long Contexts【34】
  • Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection【384】
  • Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity【385】

Chapter 11: Retrieval Paradigms—Sparse, Dense, and Late Interaction

  • Lucene 9.12 BM25Similarity: term-frequency saturation, length normalization, and IDF implementation【386】
  • Dense Passage Retrieval for Open-Domain Question Answering【349】
  • SPLADE: Sparse Lexical and Expansion Model for First Stage Ranking【387】
  • Elastic: official ELSER documentation【388】
  • ColBERTv2: Effective and Efficient Retrieval via Lightweight Late Interaction【371】
  • ColPali: Efficient Document Retrieval with Vision Language Models【358】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】

Chapter 12: Query Understanding and Rewriting

  • Precise Zero-Shot Dense Retrieval without Relevance Labels【389】
  • Take a Step Back: Evoking Reasoning via Abstraction in Large Language Models【390】
  • Query Rewriting for Retrieval-Augmented Large Language Models【391】
  • Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity【385】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】

Chapter 13: Multi-Path Retrieval, RRF Fusion, and Reranking

  • Reciprocal Rank Fusion outperforms Condorcet and individual Rank Learning Methods (SIGIR 2009)【392】
  • Lucene 9.12: the BM25Similarity scoring implementation【386】
  • Anthropic: Introducing Contextual Retrieval【352】
  • ColBERTv2: Effective and Efficient Retrieval via Lightweight Late Interaction【371】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】
  • Retrieval-Augmented Generation for Large Language Models: A Survey【350】

Chapter 14: A Five-Layer Framework for RAG Optimization

  • Searching for Best Practices in Retrieval-Augmented Generation【351】
  • Retrieval-Augmented Generation for Large Language Models: A Survey【350】
  • Anthropic: Introducing Contextual Retrieval【352】
  • Evaluation of Retrieval-Augmented Generation: A Survey【393】

Chapter 15: Advanced RAG Paradigms

  • Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection【384】
  • Corrective Retrieval Augmented Generation【394】
  • RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval【365】
  • Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity【385】
  • Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG【395】
  • Active Retrieval Augmented Generation【396】
  • Retrieval-Augmented Generation for Large Language Models: A Survey【350】

Source checks for this translation covered the original papers’ reflection tokens, corrective branches, RAPTOR soft clustering and retrieval strategies, Adaptive-RAG routing, and the survey’s three organizational categories. The Agentic RAG survey and Active Retrieval Augmented Generation were checked at the abstract level only; their experiments were not reproduced.

Chapter 16: GraphRAG and Graph Retrieval

  • From Local to Global: A Graph RAG Approach to Query-Focused Summarization【397】
  • Microsoft GraphRAG documentation【398】
  • Microsoft GraphRAG: Indexing Methods【399】
  • Microsoft GraphRAG: CLI Reference, including update methods【400】
  • Microsoft GraphRAG: Query Modes【401】
  • Microsoft GraphRAG: Local Search【402】
  • LightRAG: Simple and Fast Retrieval-Augmented Generation【403】
  • HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models【404】
  • PathRAG: Pruning Graph-based Retrieval Augmented Generation with Relational Paths【405】
  • Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG【395】

The Chinese source records access to Microsoft’s online documentation on 2026-09-15. CLI options and library implementations can change; use a pinned version and its configuration when reproducing behavior.

For this translation, the linked official documentation was checked for Standard/Fast extraction, English-oriented NLP defaults, update methods, and Local/Global/DRIFT/Basic behavior. Relevant passages in the GraphRAG, LightRAG, HippoRAG, and PathRAG papers were also checked; the Agentic RAG survey was checked at the abstract level only. No indexing runs, update tests, or comparative benchmarks were performed.

Chapter 17: Generation, Grounding, and Hallucination Mitigation

  • Astute RAG: Overcoming Imperfect Retrieval Augmentation and Knowledge Conflicts for Large Language Models【406】
  • Making Retrieval-Augmented Language Models Robust to Irrelevant Context【407】
  • RAGAS: Automated Evaluation of Retrieval Augmented Generation【408】
  • ALCE: Enabling Large Language Models to Generate Text with Citations【409】
  • Corrective Retrieval Augmented Generation【394】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】

Chapter 18: A Framework for Evaluating RAG

  • Ragas: Automated Evaluation of Retrieval Augmented Generation【408】
  • Ragas: Context Recall【410】
  • Ragas: Context Precision【411】
  • Ragas: Faithfulness【412】
  • Ragas: Answer / Response Relevancy【413】
  • ALCE: Enabling Large Language Models to Generate Text with Citations【409】
  • Evaluation of Retrieval-Augmented Generation: A Survey【393】
  • CRAG - Comprehensive RAG Benchmark【414】
  • TREC RAG Track【415】
  • Fact, Fetch, and Reason: A Unified Evaluation of Retrieval-Augmented Generation【416】

The Ragas online metric documentation was consulted on 2026-09-15. API names here distinguish measurement definitions; they do not imply that every historical version provides classes with those names.

Chapter 19: Dynamic Knowledge Base Updates and Incremental Indexing

  • FreshDiskANN: A Fast and Accurate Graph-Based ANN Index for Streaming Similarity Search【380】
  • Official pgvector README: updates, deletion, and VACUUM【379】
  • Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs【378】
  • ACORN: Performant and Predicate-Agnostic Search Over Vector Embeddings and Structured Data【381】
  • LightRAG: Simple and Fast Retrieval-Augmented Generation【403】
  • Retrieval-Augmented Generation for Large Language Models: A Survey【350】

Chapter 20: RAG Deployment Challenges and Security

  • PoisonedRAG: Knowledge Corruption Attacks to Retrieval-Augmented Generation of Large Language Models【417】
  • Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection【418】
  • Evaluation of Retrieval-Augmented Generation: A Survey【393】
  • Searching for Best Practices in Retrieval-Augmented Generation【351】
  • Retrieval-Augmented Generation for Large Language Models: A Survey【350】

Chapter 21: Multimodal RAG

  • Learning Transferable Visual Models From Natural Language Supervision (CLIP)【171】
  • ColPali: Efficient Document Retrieval with Vision Language Models【358】
  • RAG-Anything: All-in-One RAG Framework【419】
  • Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks【125】
  • Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection【418】

Source checks for this translation covered CLIP’s learned shared embedding space, ColPali’s page-image multi-vector representation and late interaction, and RAG-Anything’s cross-modal representation and retrieval approach. The original RAG and indirect prompt-injection papers were checked at the abstract level only. These checks do not constitute tests of media parsing, authorization, localization, or attack resistance.

Chapter 22: Text-to-SQL: From Business Questions to Verifiable Queries

  • Bojie Li, AI Agents in Depth, Chapter 5: Code as an Interaction Interface and Generating SQL Queries【420】. This chapter adopts the separation of query generation from execution; the business case, data, SQL, and diagram were designed separately.
  • ERP example at the same pinned commit: README【421】, agent.py【422】, and demo.py【423】. The book's experiment description uses PostgreSQL, while the runnable companion uses SQLite. The review here inspects source code without running the upstream program; its reported pass rates are not presented as this chapter's experimental results or customer benefits.
  • Official SQLite documentation: aggregate functions【424】, read-only URI mode【425】, authorization callbacks【426】, defenses for untrusted SQL【427】, application-defined function security【428】, and query plans【429】.
  • Official SQLite documentation: type affinity【430】 and STRICT tables and version requirements【431】.
  • Safdarian et al., “SchemaGraphSQL: Efficient Schema Linking with Pathfinding Graph Algorithms for Text-to-SQL on Large-Scale Databases”【432】, Findings of EACL 2026. This chapter uses its distinction between schema-graph pathfinding and joinability discovery, not its benchmark results as a production guarantee.
  • Nahid et al., “Rethinking Schema Linking: A Context-Aware Bidirectional Retrieval Approach for Text-to-SQL”【433】, Findings of EACL 2026. It motivates evaluating schema retrieval as a separate stage with both recall and false positives.

Sources were consulted on 2026-09-14; the pinned commit and SQLite typing, aggregation, and execution limits were rechecked on 2026-09-15. The English migration was checked on 2026-09-20; the schema-linking sources were rechecked on 2026-09-29.

Chapter 1: From Large Language Models to AI Agents

  • Anthropic: Introducing the Model Context Protocol【282】
  • OpenAI Agents SDK: Agents【437】
  • LangChain: Short-term memory【438】
  • MCP joins the Agentic AI Foundation【439】
  • Linux Foundation: Agent2Agent Protocol Project【314】
  • A2A v1.0.1 release specification: Core objects and message sending【312】
  • A2A Protocol: Core Concepts【440】
  • Anthropic: Building Effective Agents【275】

Chapter 2: Modern Agent System Architecture

  • OpenAI: Function calling—Chat Completions and Responses tool definitions, strict mode【262】
  • MCP Governance and Stewardship【441】
  • MCP joins the Agentic AI Foundation【439】
  • Microsoft Agent Framework Overview【442】
  • AutoGen Maintenance Mode【443】
  • Anthropic: Effective context engineering for AI agents【307】
  • Chroma Research: Context Rot — How Increasing Input Tokens Impacts LLM Performance【353】
  • LangChain: Context Engineering for Agents【444】
  • Drew Breunig: How Contexts Fail and How to Fix Them【445】
  • Anthropic: Equipping agents for the real world with Agent Skills【303】
  • Anthropic: Code execution with MCP【446】
  • Cloudflare: Code Mode — the better way to use MCP【447】
  • RAG-MCP: Mitigating Prompt Bloat in LLM Tool Selection via Retrieval-Augmented Generation【448】

Chapter 3: Tools, Skills, Agents, Workflows, and AGENTS.md

The Agent Skills format and host instruction-discovery behavior were checked against official pages accessed on 2026-09-15. That access date is not a format version number.

Chapter 4: Agent Design Patterns

  • ReAct: Synergizing Reasoning and Acting in Language Models【268】
  • Reflexion: Language Agents with Verbal Reinforcement Learning【451】
  • Reflexion: original experimental setup and Tables 1–2【452】
  • ReWOO: Decoupling Reasoning from Observations for Efficient Augmented Language Models【453】
  • An LLM Compiler for Parallel Function Calling【454】
  • LangChain: Planning Agents【455】
  • Anthropic: Building Effective Agents【275】 — the architectural distinction cited here comes from the December 2024 article; the live page now notes that parts of its tooling discussion have changed.

Chapter 5: Model Reasoning and Search Methods for Agents

  • Chain-of-Thought Prompting Elicits Reasoning in Large Language Models【104】
  • Large Language Models are Zero-Shot Reasoners【105】
  • Self-Consistency Improves Chain of Thought Reasoning in Language Models【72】
  • Least-to-Most Prompting Enables Complex Reasoning in Large Language Models【111】
  • Tree of Thoughts: Deliberate Problem Solving with Large Language Models【112】 — Game of 24 conditions checked against v2, §4.1 and Table 2【456】.
  • Graph of Thoughts: Solving Elaborate Problems with Large Language Models【457】
  • PAL: Program-aided Language Models【458】
  • Self-Refine: Iterative Refinement with Self-Feedback【459】
  • Plan-and-Solve Prompting【460】
  • Program of Thoughts Prompting【461】
  • Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting【462】
  • Let's Verify Step by Step【463】
  • DeepSeek-R1【301】 — the discussion refers to the January 2025 v1 technical report【52】, not an inference about a current hosted service.
  • Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters【464】
  • s1: Simple test-time scaling【115】
  • OpenAI: Reasoning best practices【109】 — vendor-specific guidance, including historical o-series examples; not a universal prompting rule.
  • OpenAI: Reasoning models【117】 — reasoning-token accounting and opaque continuation state; supported behavior depends on the model and API.
  • Amazon Bedrock: Extended thinking【119】 — returned thinking forms and tool-continuation requirements; the page distinguishes full output for Claude 3.7 Sonnet from summarized output for Claude 4.
  • Python 3.13: decimal — precision, rounding, and decimal arithmetic【465】

Chapter 6: Decomposing and Scheduling Complex Tasks

  • Anthropic: Building Effective Agents【275】
  • LangChain: Planning Agents【455】
  • LLMCompiler: An LLM Compiler for Parallel Function Calling【454】
  • ReWOO: Decoupling Reasoning from Observations for Efficient Augmented Language Models【453】
  • AWS Builders' Library: Making retries safe with idempotent APIs【466】
  • SHOP planners: the authors' project page and implementations【467】
  • Martin Kleppmann: How to do distributed locking【468】 — late writes after lease expiration and fencing, from the 2016 article.

Chapter 7: Memory Mechanisms in AI Agents

  • CoALA: Cognitive Architectures for Language Agents【469】
  • MemGPT: Towards LLMs as Operating Systems【470】
  • Generative Agents: Interactive Simulacra of Human Behavior【471】
  • Reflexion: Language Agents with Verbal Reinforcement Learning【451】
  • LangGraph: Memory overview【472】 (documentation snapshot 1fa2214【473】, used to distinguish within-thread and cross-thread scope)
  • OpenAI: Safety in building agents【474】 (cited for trust-boundary principles, not for its product-specific default model recommendations)
  • LongMemEval paper【475】 and official README snapshot 9e0b455【476】
  • LoCoMo paper【477】 and ACL 2024 release notes snapshot 9228632【478】
  • LongMemEval-V2 official README snapshot 2cc8c54【479】

Source review: 2026-09-15. Framework documentation is continuously updated; the snapshots fix the concepts and evaluation descriptions cited here, not a claim that their benchmark results were reproduced.

Chapter 8: Implementing Short- and Long-Term Agent Memory

  • CoALA: Cognitive Architectures for Language Agents【469】
  • MemGPT: Towards LLMs as Operating Systems【470】
  • Generative Agents: Interactive Simulacra of Human Behavior【471】
  • Reflexion: Language Agents with Verbal Reinforcement Learning【451】
  • LangGraph: Persistence【483】 (documentation snapshot 3edafc8【484】)
  • LangGraph: Checkpointers【485】
  • LangGraph: Memory overview【472】 (documentation snapshot 1fa2214【473】)
  • AWS Prescriptive Guidance: Transactional outbox pattern【486】 (illustrates dual writes, duplicate delivery, and ordering; it does not imply end-to-end exactly-once guarantees for arbitrary queues)
  • etcd v3.5: Transaction comparisons and revisions【487】 (a specific implementation of atomic conditional updates; check each other database's own guarantees)
  • LongMemEval official README snapshot 9e0b455【476】 (distinguishes the original release from the September 2025 cleaned version)
  • LongMemEval-V2 official README snapshot 2cc8c54【479】

Source review in the Chinese manuscript: September 15, 2026. The consistency, idempotency, and deletion designs are recommendations, not guarantees automatically provided by the cited frameworks or any vector database. Translation checks on September 20, 2026 covered the pinned LangGraph scope and persistence descriptions, the current checkpointer documentation, the AWS outbox discussion, etcd v3.5 transactions, and the pinned benchmark descriptions. Checks were limited to the relevant documentation sections; no framework deployment, API experiment, or benchmark reproduction was performed.

Chapter 9: Single-Agent and Multi-Agent Systems

  • Anthropic: Building Effective Agents【275】
  • Anthropic: How we built our multi-agent research system【488】
  • Cognition: Don't Build Multi-Agents【489】
  • Why Do Multi-Agent LLM Systems Fail? (MAST)【490】
  • Agent2Agent (A2A) v1.0.1 Specification【311】
  • Raft: Algorithm and papers maintained by the authors【491】
  • etcd v3.5: Revisions, conditional transactions, and leases【487】
  • Martin Kleppmann: How to do distributed locking (2016)【468】 (cited for its analysis of process pauses, leases, and fencing tokens, without generalizing its Redis-version conclusions to current products)
  • AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation【492】
  • CAMEL: Communicative Agents for Mind Exploration of Large Language Model Society【493】

Source review recorded in the original manuscript: 2026-09-15. The Anthropic–Cognition comparison is limited to the 2025 systems and models described in those articles; their internal evaluations were not reproduced. A2A was checked against the v1.0.1 release tag, not the stale Latest Released Version 1.0.0 wording at the top of the specification page.

Chapter 10: Agent Memory and Context Compression

  • MemGPT: Towards LLMs as Operating Systems【470】
  • Anthropic: Effective context engineering for AI agents【307】
  • Chroma Research: Context Rot【353】
  • LangChain: Context Engineering for Agents【444】
  • Anthropic Prompt Caching【494】
  • OpenAI Prompt Caching【80】
  • OpenAI Compaction【495】
  • LangGraph Memory【496】
  • LoCoMo official release notes, snapshot 9228632【478】 (ACL 2024 release; this snapshot's data/locomo10.zip contains ten per-conversation JSON files)
  • LongMemEval official documentation, snapshot 9e0b455【476】 (distinguishes the original release from the September 2025 cleaned release)
  • LongMemEval-V2 official documentation, snapshot 2cc8c54【479】

Source review recorded in the original manuscript: 2026-09-15. The OpenAI caching and compaction references are rolling Responses API documentation and do not promise identical pricing, thresholds, or history-passing rules across models or versions. The server-side APIs were not tested in that review. The Anthropic caching entry remains supplementary reading; it was not used to add conclusions about specific parameters.

Chapter 11: Giving LLMs and Agents the Ability to Plan

  • Chain-of-Thought Prompting Elicits Reasoning in Large Language Models【104】
  • Tree of Thoughts: Deliberate Problem Solving with Large Language Models【112】
  • Graph of Thoughts: Solving Elaborate Problems with Large Language Models【457】
  • Plan-and-Solve Prompting【460】
  • LLMCompiler: An LLM Compiler for Parallel Function Calling【454】
  • Anthropic: Building Effective Agents【275】
  • LLMs Can't Plan, But Can Help Planning in LLM-Modulo Frameworks【497】
  • Do As I Can, Not As I Say: Grounding Language in Robotic Affordances【498】
  • OR-Tools: CP-SAT Solver【499】
  • DeepSeek-R1【301】

Chapter 12: Agent Reflection, Verification, and Self-improvement

  • Self-Refine: Iterative Refinement with Self-Feedback【459】
  • Reflexion: Language Agents with Verbal Reinforcement Learning (v4)【500】
  • Language Agent Tree Search Unifies Reasoning, Acting, and Planning (v3)【501】
  • Improving Factuality and Reasoning in Language Models through Multiagent Debate【502】
  • Anthropic: Building Effective Agents【275】
  • Large Language Models Cannot Self-Correct Reasoning Yet【503】
  • Training Language Models to Self-Correct via Reinforcement Learning (SCoRe, v1)【504】

Chapter 13: Multi-Agent Coordination, Routing, and Dynamic Switching

The Chinese source checked framework behavior against official documentation accessible on 2026-09-15; the English review rechecked the relevant documentation on 2026-09-20. Deployment still requires a pinned SDK version. The A2A v1.0.1 release【309】 was published on 2026-05-28, although the specification page at that tag still labels v1.0.0 as the latest release. This chapter identifies the specification by its release tag, not by the page's “latest” notice.

Live framework documentation and the LangGraph source checked during translation are not snapshots of a deployed SDK. Raft's replicated-log and non-Byzantine failure assumptions were checked against the authors' overview and Ongaro's thesis source, not the linked paper PDF. Anthropic's article now notes that its tooling discussion dates from December 2024; it is used here for architectural patterns, not current interface guarantees.

  • LangGraph Graph API【506】
  • LangGraph: Concurrent State Update Error【507】
  • LangGraph Persistence【483】
  • OpenAI Agents SDK【508】
  • OpenAI Swarm: Official Guidance on Migrating to the Agents SDK【509】
  • OpenAI Agents SDK: Handoffs【510】
  • A2A v1.0.1 Protocol Specification【311】
  • Amazon SQS: At-Least-Once Delivery【511】
  • etcd v3.6: API Consistency and Lease Guarantees【512】
  • etcd v3.6: Lock and Transactional Protection【513】
  • AWS: Transactional Outbox【486】
  • Raft: Authors' Algorithm and Paper Resources【491】
  • Anthropic: Building Effective Agents【275】
  • LangGraph: add_messages Implementation【514】
  • OpenAI Agents SDK: Manager and Handoff Patterns【437】
  • Diego Ongaro's Thesis: Replicated State Machines and Failure Assumptions【515】

Chapter 14: Agent Evaluation and Benchmarks

  • SWE-bench: Can Language Models Resolve Real-World GitHub Issues?【154】
  • OpenAI: Why SWE-bench Verified no longer measures frontier coding capabilities【155】
  • SWE-Lancer: Can Frontier LLMs Earn $1 Million from Real-World Freelance Software Engineering?【516】
  • Terminal-Bench: Official tasks and versions【517】
  • GAIA: a benchmark for General AI Assistants【518】
  • tau-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains【162】
  • tau^2-Bench: Evaluating Conversational Agents in a Dual-Control Environment【519】
  • OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments【218】
  • WebArena: A Realistic Web Environment for Building Autonomous Agents【219】
  • BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents【520】
  • AgentBench: Evaluating LLMs as Agents【521】
  • MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering【522】
  • Humanity's Last Exam【165】
  • AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents【523】
  • OpenTelemetry: Generative AI Semantic Conventions【524】
  • OpenTelemetry: GenAI convention status (Development, pinned commit 0c87594)【525】
  • Anthropic: Building Effective Agents【275】
  • Anthropic: Demystifying evals for AI agents【526】
  • SWE-bench: Evaluation harness【527】
  • OpenAI: BrowseComp reference implementation【528】
  • Anthropic: How we built our multi-agent research system【488】

Chapter 15: Agent Security and Prompt Injection

  • Simon Willison: The lethal trifecta for AI agents【530】
  • Simon Willison: The Dual LLM pattern for building AI assistants that can resist prompt injection【531】
  • Design Patterns for Securing LLM Agents against Prompt Injections (v1)【532】
  • Defeating Prompt Injections by Design (CaMeL, v2)【533】
  • Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection【418】
  • AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents【523】
  • The Task Shield: Enforcing Task Alignment to Defend Against Indirect Prompt Injection in LLM Agents【534】
  • System-Level Defense against Indirect Prompt Injection Attacks: An Information Flow Control Perspective【535】
  • Bypassing LLM Guardrails: An Empirical Analysis of Evasion Attacks against Prompt Injection and Jailbreak Detection Systems (v3)【536】
  • OWASP Top 10 for LLM Applications【341】
  • Docker Engine: seccomp configuration and boundaries for Linux containers【537】
  • Anthropic: Equipping agents for the real world with Agent Skills【303】
  • MCP 2026-07-28: Security Best Practices【343】

Chapter 16: Agent Harnesses—Definitions, Boundaries, and Layers

  • Anthropic: Building Effective Agents【275】
  • SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering【538】
  • METR: Measuring AI Ability to Complete Long Software Tasks【539】
  • Simon Willison: Designing agentic loops【540】
  • Claude Agent SDK: Overview【541】
  • GitHub Docs: Configure the development environment for Copilot cloud agent【542】
  • LangGraph: Persistence【483】
  • Microsoft Agent Framework Overview【442】

Chapter 17: The Agent Loop and Runtime State Machine

Chapter 18: Context Engineering and Context Assembly

  • Anthropic: Effective context engineering for AI agents【307】
  • Model Context Protocol: Tools【285】
  • Claude Agent SDK: Modifying system prompts【548】
  • OpenAI: Prompt caching【80】: prefix matching, model differences, and cache breakpoints; accessed in the source review on 2026-09-15.
  • Anthropic Prompt Caching【494】: retained as the original technical source. Access during the source review redirected to a region-unavailable page, so it was not used to confirm current parameters.
  • LangChain: Context Engineering for Agents【444】

Chapter 19: Tool Registries, Call Contracts, and Execution Pipelines

  • Model Context Protocol: Tools【285】
  • Model Context Protocol: Versioning【276】: the Current status and revision rules for 2026-07-28; accessed in the source review on 2026-09-15.
  • OpenAI Agents SDK: Guardrails【549】
  • Anthropic: Code execution with MCP【446】
  • Cloudflare: Code Mode — the better way to use MCP【447】
  • RAG-MCP: Mitigating Prompt Bloat in LLM Tool Selection via Retrieval-Augmented Generation【448】

Chapter 20: Permissions, Sandboxing, and Isolation

  • Claude Agent SDK: Configure permissions【550】
  • gVisor documentation【551】
  • GitHub Docs: Configure the development environment for Copilot cloud agent【542】
  • GitHub Docs: Customizing or disabling the firewall【552】
  • Simon Willison: Designing agentic loops【540】
  • Simon Willison: The lethal trifecta for AI agents【530】

The source review checked SDK permission order and GitHub environment/firewall coverage against official documentation on 2026-09-15. The v2.1.199 requirement for anthropic/requiresUserInteraction is retained. No SDK permission experiments or cloud-environment configuration were performed in that review.

Chapter 21: Checkpoints, Persistence, Retries, Timeouts, Idempotency, and Recovery

Chapter 22: Human-in-the-Loop Approval and Interruptible Execution

Chapter 23: Tracing, Evaluation, Cost Control, and Coding-Agent Harness Case Studies

  • OpenTelemetry: Generative AI Semantic Conventions【524】
  • OpenTelemetry: Traces【557】: spans, span events, context propagation, and span links.
  • Claude Agent SDK: How the agent loop works【545】
  • OpenAI Agents SDK: Running agents【546】
  • OpenAI Agents SDK: Guardrails【549】
  • OpenAI Codex repository【558】
  • GitHub Docs: Configure the development environment for Copilot cloud agent【542】
  • Simon Willison: Designing agentic loops【540】
  • Anthropic: How we built our multi-agent research system【488】

The GenAI semantic conventions are pinned to commit 0c87594975195608dc91b3f702e250a7b240c151, accessed in the source review on 2026-09-15. Product claims are bounded by the official documentation cited here; they do not imply knowledge of unpublished internal implementations.

Chapter 24: Code Search, Editing, and Verification for Coding Agents

  • Bojie Li, Understanding AI Agents: Design Principles and Engineering Practice, relevant Chapter 5 passages【559】: search categories, edit formats, and immediate feedback. Pinned commit 985a49d35b9f50937f1f757cf25867672991ded7; consulted 2026-09-14.
  • edit_tool.py at the same commit【560】: matching and post-write checking, especially the write-then-check sequence without automatic rollback.
  • Language Server Protocol 3.17【561】: definitions, references, call hierarchies, and position encodings. Consulted 2026-09-14.
  • Python ast.parse documentation【562】: the boundary between AST parsing and complete compilation checks. Consulted 2026-09-14.

Chapter 25: Agent Post-Training: From Failed Trajectories to Reliable Policies

  • Bojie Li, ai-agent-book, Chapter 7: failure attribution, exact-copy diagnosis, and trajectory-prefix regression【563】.
  • Bojie Li, Chapter 8: multi-turn credit assignment, rewards, and on-policy distillation【564】, and from failure cases to post-training【565】.
  • Bojie Li, companion experiment: README for the premature-completion DPO repair【566】, particularly its data, metric definitions, and trust-boundary discussion.

Sources were consulted on 2026-09-14 at pinned commit 985a49d35b9f50937f1f757cf25867672991ded7. DPO results are reported from the README; training was not independently reproduced. The difference between candidate comparisons and free generation cannot be extrapolated to improved overall task completion. References to methods such as RLVP do not mean that this chapter carried out those training experiments.

Chapter 1: An Overview of Major AI Agent Development Frameworks

Chapter 2: Chain Design and LCEL

Chapter 3: The Underlying Architecture of LangChain v1

Chapter 4: Seven Steps to a Production-Ready LangChain Agent

Chapter 5: Tool Registration and Tool Contracts

Chapter 6: Short-Term and Long-Term Memory in LangChain

Chapter 7: Dividing Responsibilities Between LangChain and LlamaIndex

Chapter 8: LangChain4j and the Java Ecosystem

Chapter 9: The Layered Relationship Between LangChain and LangGraph

Chapter 10: LangGraph's Core Advantages

Chapter 11: LangChain's Version Evolution

Chapter 12: How Deep Research Works

Chapter 13: Building a Production Quality Feedback Loop with LangSmith

Chapter 14: LlamaIndex Data and Index Abstractions

  • LlamaIndex official documentation【623】
  • LlamaIndex: Loading Data (Ingestion Pipeline)【624】
  • LlamaIndex: Indexing concepts【625】
  • LlamaIndex: Default and optional retrieval modes for each index【626】
  • LlamaIndex: Property Graph Index【627】
  • LlamaIndex: Storage concepts【628】

Chapter 15: LlamaIndex Query Engines and Workflow Orchestration

Chapter 16: DSPy's Declarative Programming Model: Signatures, Modules, and Programs

  • Official DSPy documentation【637】
  • DSPy: Class-based signatures【638】
  • DSPy: Changing modules【639】
  • DSPy paper: Khattab et al., "DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines"【640】

Chapter 17: DSPy's Compiler and Optimizers

  • DSPy: Choosing an optimizer【641】
  • DSPy: MIPROv2 API and its three-stage mechanism【642】
  • DSPy: GEPA API and feedback contract【643】
  • DSPy: GEPA optimization tutorial【644】
  • DSPy paper: Khattab et al., "DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines"【640】
  • GEPA paper: Agrawal et al., "GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning"【645】
  • Official LangSmith documentation【646】

Chapter 18: Semantic Kernel's Core Abstractions: Kernel, Plugin, and Planner

  • Semantic Kernel official documentation: Introduction【647】
  • Semantic Kernel: Kernel concepts【648】
  • Semantic Kernel: Plugin concepts【649】
  • Semantic Kernel: Planning concepts【650】
  • Semantic Kernel: Filter concepts【651】
  • Semantic Kernel official repository and successor-framework announcement, pinned commit【652】

Version note: The MAF successor relationship and 1.0 release statement were checked on 2026-09-15. This statement does not cover every language or experimental extension; Chapter 19 explains the boundaries.

Chapter 19: Semantic Kernel's Process Framework and Agent Framework

  • Semantic Kernel: Process Framework【653】
  • Semantic Kernel: Agent Framework overview【654】
  • Semantic Kernel: Agent Orchestration【655】
  • Semantic Kernel: Complete Group Chat runtime example【656】
  • Semantic Kernel official repository's MAF 1.0 statement, pinned commit【652】
  • Microsoft Agent Framework overview【442】
  • SK → MAF migration guide【657】
  • Microsoft: Semantic Kernel and MAF support transition announcement【658】
  • LangGraph official documentation【568】

Version note: MAF 1.0, the Go public preview, the experimental labels for SK Process/Orchestration, and the transition announcement were checked on 2026-09-15. The transition announcement retains its original description of MAF as still in Preview; it must not override the later 1.0 release statement or be used to infer an exact SK EOL date.

Chapter 20: Multi-Agent Abstractions in AutoGen and CrewAI

Version note: AutoGen's maintenance mode and successor recommendation were checked against the pinned commit above on 2026-09-15. The code in this chapter retains the Core/AgentChat generation of APIs rather than rewriting it into MAF APIs.

Chapter 21: PydanticAI's Type-Safe Approach and Where the Three Frameworks Fit

Chapter 22: Comparing Framework Internals: State, Persistence, Tool Contracts, and Observability

Version note: see the fixed sources in Chapters 19 and 20 for the release and maintenance status of Microsoft's frameworks. The OpenTelemetry GenAI Development status was checked on 2026-09-15. Pin and verify the semantic-convention version used by each exporter.

Chapter 23: Identifying Lock-in, Designing Portable Architectures, and Planning Migrations

  • LangGraph: Persistence【483】
  • Semantic Kernel: Process Framework【653】
  • Microsoft Agent Framework overview and successor relationship【442】
  • SK → MAF migration guide【657】
  • AutoGen official maintenance-mode notice【443】
  • PydanticAI: Durable Execution【673】
  • OpenTelemetry Generative AI semantic-conventions repository【524】
  • Martin Fowler: StranglerFigApplication【675】
  • Martin Fowler: CanaryRelease【676】
  • Alistair Cockburn: Hexagonal Architecture【677】

Chapter 1: What Is LLMOps? How It Relates to MLOps and DevOps

  • Google Cloud: MLOps: Continuous delivery and automation pipelines in machine learning【678】
  • a16z: What Is LLMOps?【679】
  • Chip Huyen: Building LLM applications for production【680】
  • OpenAI: Best practices for production deployments【681】
  • Anthropic: Building effective agents【275】

Chapter 2: A Production Architecture for AI Applications

  • OpenAI: Production best practices【681】
  • Anthropic: Building effective agents【275】
  • Google SRE Book: Chapter 1 - Introduction【682】
  • Uber Engineering: Michelangelo Machine Learning Platform【683】
  • Martin Fowler: Continuous Delivery for Machine Learning【684】

Chapter 3: Model Gateways, Routing, and Fallback

  • LiteLLM: Routing【685】
  • OpenAI: Reasoning models【117】
  • Amazon Bedrock: Model routing (intelligent prompt routing)【686】
  • Martin Fowler: CanaryRelease【676】
  • Netflix Tech Blog: Fault Tolerance in a High Volume, Distributed System【687】

Chapter 4: Retries, Timeouts, Idempotency, and Circuit Breakers

  • Google Cloud: Implementing exponential backoff【688】
  • Stripe API: Idempotent requests【556】
  • Martin Fowler: CircuitBreaker【689】
  • Netflix Tech Blog: Fault Tolerance in a High Volume, Distributed System【687】
  • AWS Well-Architected Framework: REL05-BP04 Bulkhead architecture【690】

Chapter 5: Structured Outputs and Contract Validation

Chapter 6: Guardrails and Graceful Degradation

Chapter 7: Offline Evaluation and Eval-Driven Development

A tool’s lifecycle is not the lifecycle of an evaluation method. OpenAI’s announcement of 2026-06-03 states that its hosted Evals platform will become read-only on 2026-10-31, with the Evals dashboard and API scheduled to shut down on 2026-11-30. Teams using that platform should check their migration plans; this does not imply that the open-source openai/evals project or self-hosted evaluation methods also stop being valid.

  • OpenAI Evals【698】
  • OpenAI: Evaluation best practices【699】
  • OpenAI: 2026-06-03 Evals platform deprecation【700】
  • Anthropic: Demystifying evals for AI agents【526】
  • SciPy: Binomial proportion confidence intervals【701】
  • Google: Rules of Machine Learning - Rule #4: Keep the first model simple and get the infrastructure right【702】
  • Braintrust: What is an eval?【703】
  • LangSmith: Evaluation concepts【704】

Chapter 8: Production Observability and Tracing

The original GenAI documentation has moved to a dedicated repository, and the old location is no longer maintained. The Chinese source manuscript records a check of the migration notice on 2026-09-15.

  • OpenTelemetry: Semantic conventions for generative AI systems【705】
  • OpenTelemetry: GenAI semantic conventions repository【524】
  • Google SRE Book: Monitoring Distributed Systems【706】
  • LangSmith Observability【600】
  • Arize Phoenix: Tracing【707】
  • Honeycomb: Observability for LLMs【708】

Chapter 9: Prompt, Model, and Data Versioning

Chapter 10: LLM CI/CD, Staged Rollouts, Canary Releases, and A/B Tests

  • Martin Fowler: CanaryRelease【676】
  • Google SRE Workbook: Canarying Releases【713】
  • SciPy: ttest_ind, independent samples, and Welch’s test assumptions【714】
  • Martin Fowler: Continuous Delivery for Machine Learning【684】
  • Spinnaker: Canary Analysis【715】
  • Optimizely: Statistical significance in A/B testing【716】

Chapter 11: Caching, Batching, Throughput, and Cost Optimization

Chapter 12: SLOs, Capacity Planning, and Incident Response

Chapter 13: Feedback Loops and the Data Flywheel

  • OpenAI: Fine-tuning【726】
  • Anthropic: Constitutional AI and RLHF【727】
  • LangSmith: Attach user feedback【620】
  • Netflix Tech Blog: Recommendation systems and the data flywheel【728】
  • Google: People + AI Guidebook - Feedback + Control【729】

Chapter 1: AI System Threat Modeling and the Attack Surface

  • MITRE ATLAS【730】
  • NIST AI Risk Management Framework (AI RMF 1.0)【176】
  • NIST Generative AI Profile (NIST AI 600-1)【731】
  • OWASP Top 10 for Large Language Model Applications【341】
  • OWASP Agentic AI Threats and Mitigations【732】
  • OWASP Top 10 for Agentic Applications 2026【733】
  • NIST AI 100-2 E2025: Adversarial Machine Learning Taxonomy【734】

Chapter 2: Prompt Injection, Jailbreaks, and System Prompt Leakage

  • OWASP LLM01:2025 Prompt Injection【735】
  • OWASP LLM07:2025 System Prompt Leakage【736】
  • Many-shot Jailbreaking (Anthropic)【737】
  • Universal and Transferable Adversarial Attacks on Aligned Language Models【738】
  • Ignore This Title and HackAPrompt: Exposing Systemic Vulnerabilities of LLMs【739】
  • MITRE ATLAS: Prompt Injection【740】

Chapter 3: Output Handling and Secret Exfiltration

  • OWASP LLM05:2025 Improper Output Handling【741】
  • OWASP LLM02:2025 Sensitive Information Disclosure【742】
  • Imprompter: Tricking LLM Agents into Improper Tool Use【743】
  • Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection【418】
  • OWASP: CSV Injection【744】

Chapter 4: Training, Fine-Tuning, and RAG Data Poisoning

  • Poisoning Web-Scale Training Datasets is Practical【745】
  • Anthropic / UK AISI / Alan Turing Institute: Small samples can poison language models【746】
  • BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chain【747】
  • A Backdoor Attack Against LSTM-based Text Classification Systems【748】
  • On the Exploitability of Instruction Tuning【749】
  • OWASP LLM04:2025 Data and Model Poisoning【750】
  • NIST AI 100-2 E2025: Adversarial Machine Learning Taxonomy【734】

Chapter 5: Model Supply Chains and Serialization Risks

  • OWASP LLM03:2025 Supply Chain【751】
  • Hugging Face: Pickle Scanning and Safetensors【752】
  • PyTorch 2.6: Serialization semantics and weights_only【753】
  • Sleepy Pickle: Exploiting Machine Learning Pickle Files【754】
  • MITRE ATLAS: ML Supply Chain Compromise【755】
  • CycloneDX: Machine Learning Bill of Materials (ML-BOM)【756】
  • Sigstore: Software Signing for Everyone【757】

Chapter 6: Privacy, PII, and Memory Leakage

  • Extracting Training Data from Large Language Models【758】
  • Quantifying Memorization Across Neural Language Models【759】
  • Membership Inference Attacks against Machine Learning Models【760】
  • Deep Learning with Differential Privacy (DP-SGD)【761】
  • OWASP LLM02:2025 Sensitive Information Disclosure【742】
  • NIST AI 600-1: Generative AI Profile — Privacy risks【731】
  • OpenAI: Data controls in the API platform【170】
  • GDPR original text: Articles 5, 6, 17, and Chapter V【762】

Chapter 7: Least Privilege and Identity Governance for Agents, Tools, MCP, and A2A

  • Confused Deputy Problem (Norm Hardy, 1988)【763】
  • MCP 2026-07-28 Authorization: Confused Deputy Considerations【342】
  • RFC 8693: OAuth 2.0 Token Exchange, especially Sections 1.1 and 4.1【764】
  • RFC 7662: OAuth 2.0 Token Introspection【765】
  • OWASP Agentic AI Threats and Mitigations: Identity and Authorization【732】
  • NIST SP 800-207: Zero Trust Architecture【766】
  • SPIFFE/SPIRE: Workload Identity Framework【767】

Chapter 8: Sandboxing and Network Isolation for Code Execution, Browsers, and Computer Use

  • OWASP LLM06:2025 Excessive Agency【768】
  • gVisor: Application Kernel for Containers【769】
  • Firecracker: Secure and Fast microVMs【770】
  • MDN: Same-origin policy【771】
  • Anthropic: Computer Use Demo — security precautions and isolation limitations【772】
  • MITRE ATLAS: Evade ML Model / LLM Prompt Injection【740】

Chapter 9: Security Evaluation, Red Teaming, and Continuous Assurance

  • NIST AI RMF: Measure Function【176】
  • garak: LLM Vulnerability Scanner【773】
  • PyRIT: Python Risk Identification Tool for generative AI【774】
  • HarmBench: A Standardized Evaluation Framework for Automated Red Teaming【775】
  • Red Teaming Language Models with Language Models【776】
  • OWASP LLM Applications Cybersecurity and Governance Checklist【777】

Chapter 10: AI Governance, Risk Classification, Auditing, and Content Provenance

Regulatory review date: 2026-09-15. Regulation 2026/1744 was published in the Official Journal of the European Union on 2026-07-24 and, under Article 4, entered into force on the third day after publication, 2026-07-27. Article 1, point (40), amends Article 113 of the AI Act; point (39) concerns transitional arrangements for existing systems. The discussion above cites an amendment already in force, not a proposal or political agreement. The description of C2PA's capabilities uses version 2.2 below and does not claim that it is the latest version.

  • NIST AI RMF: Govern Function【176】
  • EU AI Act: original text of Regulation (EU) 2024/1689【778】
  • Regulation (EU) 2026/1744: Article 1, points (39) and (40), and Article 4【779】
  • GDPR: Articles 33 and 34, breach notification conditions and deadlines【762】
  • European Commission: AI Act policy overview【780】
  • C2PA 2.2: Explainer, verification capabilities and non-goals【781】
  • Model Cards for Model Reporting【782】
  • System Cards: A New Resource for Understanding How AI Systems Work【783】
  • OWASP LLM Applications Cybersecurity and Governance Checklist【777】

Chapter 1: Forward Deployed Engineering: From Business Problems to Reusable Production Systems

The original source collection was compiled on 2026-09-08. At that time, three Palantir Medium/engineering-blog links returned 403; they remain further reading, and the chapter does not rely on their project details. The OpenAI Gov job link is only a role entry point.

The Evals retirement schedule, API data controls, and the specific Microsoft, Baseten, and AWS/INRIX case descriptions were rechecked on 2026-09-15. OpenAI's notice still listed existing evals becoming read-only on 2026-10-31 and the dashboard and API scheduled to shut down on 2026-11-30. Evaluation methods in older guides should be distinguished from instructions for the hosted platform; this does not invalidate open-source evaluation methods.

  • Palantir: Dev versus Delta—Demystifying Engineering Roles at Palantir【784】
  • Palantir: A Day in the Life of a Forward Deployed Software Engineer【785】
  • Palantir: AIP Chatbot Studio Overview【786】
  • Palantir: Security and Governance【787】
  • OpenAI Deployment Company: Build, Prove, Generalize【788】
  • OpenAI: Forward Deployed Engineer, Gov【789】
  • OpenAI: Evaluation Best Practices—Distinguish Methods from the Hosted Platform【699】
  • OpenAI: Deprecations, Including the Evals Platform Schedule【710】
  • OpenAI: Production Best Practices【790】
  • OpenAI: Data Controls【170】
  • Anthropic: Building Effective AI Agents【275】
  • Anthropic: How We Built Claude Managed Agents【791】
  • Palantir: Securing Software at the Speed of AI—Official Engineering Case, 2026【792】
  • Anthropic: Demystifying Evals for AI Agents—Official Engineering Article【526】
  • Microsoft: Only Believe What You Can Validate—Customer Field Engineering Account, 2026【793】
  • Baseten: Forward Deployed Engineering on the Frontier of AI—Official Team Practices, 2025【794】
  • AWS and INRIX: Transportation-Planning PoC—Coauthored by Customer and Vendor, 2025【795】
  • @vasuman: Forward Deployed Engineering 101—Practitioner's Long-Form Article on X【796】
  • Hamel Husain: A Field Guide to Rapidly Improving AI Products【797】
  • Varick Agents: Careers【798】
  • run_maotui: Working with AI in Product Management【799】 (reference for Section 1.2.1: verifying research, choosing requirements, handing off PRDs, and maintaining project decisions; accessed 2026-09-17)

Chapter 2: FDE Delivery Pitfalls and Lessons

Sources were compiled on 2026-09-10. On 2026-09-15, OpenFDE's evaluation records, task-acceptance conditions, and Claude extraction data flow were additionally checked at the cited commit. This chapter adapts working methods to the order scenario; it does not treat any project's complete process as a universal standard. Links are pinned to the commits read so readers can compare them.

Project Suggested starting point Cited version
Applied AI Field Guide【800】 Five-minute introduction【801】, scope changes and handoff exercises in the invoice case 6b557eb
FDEstack【802】 Overview of eight skills【803】, especially the connection between unresolved questions, PoC findings, and production implementation 5524460
OpenFDE【804】 Architecture【805】: how sources, facts, tasks, and context connect e2dec16
FDEOps【806】 Project overview【807】, requirement-change and handoff skills cc96340

For technical implementation, continue with Retries and Idempotency, Versioning, and Feedback Data Processing.