Unlocking Data with Generative AI and RAG, Second Edition — 书籍拆解

读到哪:未读。 readState 不是 read/partial 的书不能当锚。

项
作者	Keith Bourne
版次	2025 版
格式	epub ｜ 文本源 epub-builtin
许可	自购/个人收藏
来源	主人个人藏书,2026-08 放入收件箱
清洗	删页眉页脚 0 行、页码 0 行、断词接回 0 处

我们重写的拆解(0 章)

（还没写。拆解是这本书对我们的真正产出——底下的元数据只是索引。）

为什么收它

RAG 基础概念的备查出处。

合法性

自购/个人收藏。来源:主人个人藏书,2026-08 放入收件箱。原始文件不入库,转码文本入库(私有库)。

出版方怎么说

（起草参考,不是我们的判断。真正的「覆盖什么/不覆盖什么」写进 frontmatter 的 claims / notCovered)

出版方简介(仅供起草参考,不是我们的判断):This hands-on guide explores how to design AI agents powered by Retrieval-Augmented Generation (RAG), with cutting-edge coverage of memory systems, LangMem, and GraphRAG

它覆盖什么、不覆盖什么

（还没读到能下判断的程度。claims / notCovered 空着就是空着,不猜。)

怎么引用它

(依据: book=unlocking-data-genai-rag §What Is Retrieval-Augmented Generation?)

章节名对不上会被 lab:validate 拦下;页码锚(§p.123)同样可用。

结构(156 段,共 1114k 字符)

段	章节	页	规模
01	Contributors	—	5.5k
02	Preface	—	30.0k
03	Part 1	—	0.9k
04	What Is Retrieval-Augmented Generation?	—	45.3k
05	Code Lab: An Entire RAG Pipeline	—	50.8k
06	Practical Applications of RAG	—	39.0k
07	Components of a RAG System	—	31.8k
08	Managing Security in RAG Applications	—	48.2k
09	Part 2	—	0.9k
10	Interfacing with RAG and Gradio	—	21.3k
11	The Key Role Vectors and Vector Stores Play in RAG	—	2.9k
12	What is the difference between embeddings and vectors?	—	0.9k
13	What is a vector?	—	1.7k
14	Vector dimensions and size	—	8.8k
15	Vectorization occurs in two places	—	1.0k
16	Vector databases/stores store and contain vectors	—	1.2k
17	Vector similarity compares your vectors	—	10.9k
18	Term frequency-inverse document frequency (TF-IDF)	—	8.1k
19	Word2Vec, Sentence2Vec, and Doc2Vec	—	5.8k
20	Bidirectional encoder representations from transformers	—	5.0k
21	OpenAI and other similar large-scale embedding services	—	5.6k
22	Quality of the embedding	—	1.2k
23	Cost	—	0.8k
24	Network availability	—	1.3k
25	Speed	—	0.9k
26	Embedding compatibility	—	3.0k
27	Data sources (other than vector)	—	4.1k
28	Vector stores	—	4.5k
29	Common vector store options	—	16.2k
30	Get This Book’s PDF Version and Exclusive Extras	—	0.3k
31	Similarity Searching with Vectors	—	7.0k
32	Semantic search example	—	6.1k
33	Euclidean distance (L2)	—	2.3k
34	Dot product (also called inner product)	—	1.8k
35	Cosine distance	—	3.4k
36	Dense search	—	1.0k
37	Sparse search	—	1.6k
38	Hybrid search	—	31.4k
39	k-NN	—	2.6k
40	ANN	—	13.2k
41	pgvector	—	1.8k
42	Elasticsearch	—	1.0k
43	FAISS	—	0.9k
44	Google Vertex AI Vector Search	—	1.3k
45	Azure AI Search	—	1.3k
46	Approximate Nearest Neighbors Oh Yeah	—	0.9k
47	Pinecone	—	1.1k
48	Weaviate	—	1.1k
49	Chroma	—	2.8k
50	Evaluating RAG Quantitatively and with Visualizations	—	8.8k
51	Embedding model benchmarks	—	1.7k
52	Vector store and vector search benchmarks	—	1.9k
53	LLM benchmarks	—	1.9k
54	Final thoughts on standardized evaluation frameworks	—	2.3k
55	How do you use the ground truth?	—	0.4k
56	Generating the ground truth	—	0.3k
57	Human annotation	—	0.2k
58	Expert knowledge	—	1.2k
59	Crowdsourcing	—	0.3k
60	Synthetic ground truth	—	6.2k
61	Setting up LLMs/embedding models	—	6.9k
62	Generating the synthetic ground truth	—	7.8k
63	Analyzing the ragas results	—	6.9k
64	Retrieval evaluation	—	3.7k
65	Generation evaluation	—	6.5k
66	Ragas founder insights	—	2.7k
67	Bilingual Evaluation Understudy (BLEU)	—	0.7k
68	Recall-Oriented Understudy for Gisting Evaluation (ROUGE)	—	0.7k
69	Semantic similarity	—	0.8k
70	Human evaluation	—	4.0k
71	Get This Book’s PDF Version and Exclusive Extras	—	0.3k
72	Key RAG Components in LangChain	—	55.6k
73	Using LangChain to Get More from RAG	—	36.0k
74	Part 3	—	1.8k
75	Combining RAG with the Power of AI Agents and LangGraph	—	49.5k
76	Ontology-Based Knowledge Engineering for Graphs	—	43.3k
77	Graph-Based RAG	—	11.1k
78	Step 1 – a recap: what you have in Protégé	—	0.6k
79	Step 2 – preparing your notebook environment	—	3.1k
80	Step 3 – converting your Protégé ontology for Neo4j import	—	8.4k
81	Step 4 – importing nodes, edges, and data properties into Neo4j	—	9.1k
82	Step 5 – add navigational anchor nodes (stocks, bonds, and so on)	—	3.1k
83	Step 5.2 – create All X concept nodes and wire members	—	3.5k
84	Step 6 – enable hybrid embeddings (text + structure) and multi-hop support	—	14.0k
85	Step 7 – vector search and graph expansion (ready-to-prompt context)	—	10.5k
86	Step 8 – generate with LangChain and OpenAI	—	3.6k
87	Best practices and next steps	—	3.8k
88	Semantic Caches	—	10.9k
89	Core components of semantic caches	—	2.7k
90	The intelligence layer between the query and the response	—	11.7k
91	Step 1 – installing dependencies	—	0.9k
92	Step 2 – setting up	—	1.7k
93	Step 3 – basic semantic cache	—	4.9k
94	Step 4 – entity masking for better generalization	—	4.6k
95	Step 5 – cross-encoder verification	—	5.2k
96	Step 6 – adaptive thresholds	—	4.7k
97	Step 7 – auto-population with fallback	—	6.0k
98	Putting it all together	—	2.5k
99	LLM-based paraphrasing	—	0.5k
100	Back-translation for natural variation	—	1.0k
101	Synonym and lexical expansion	—	1.2k
102	Synthetic query generation	—	1.8k
103	Domain-specific constraints	—	3.5k
104	Time-based eviction	—	0.9k
105	Least recently used with semantic decay	—	1.4k
106	Performance-based pruning	—	1.8k
107	Semantic clustering for redundancy elimination	—	3.2k
108	Get This Book’s PDF Version and Exclusive Extras	—	0.3k
109	Agentic Memory: Extending RAG with Stateful Intelligence	—	5.5k
110	The pre-ChatGPT era – state machines and slot filling	—	14.3k
111	Working memory (short-term)	—	2.4k
112	Episodic memory (experiences/events)	—	1.7k
113	Semantic memory (facts/knowledge)	—	1.7k
114	Procedural memory (skills)	—	4.1k
115	Community/public memory – shared across all users	—	2.0k
116	Personal/user-specific memory – individual context	—	2.5k
117	How memory types and scopes intersect	—	5.3k
118	Quality of working memory data	—	1.0k
119	Collecting comprehensive user experience events	—	1.0k
120	Curation and maintenance of growing memory stores	—	2.6k
121	Mem0 architecture	—	2.0k
122	LangMem	—	2.1k
123	Zep and Graphiti	—	2.8k
124	Choosing the right memory framework	—	5.2k
125	Evaluating memory-based versus memoryless systems	—	2.7k
126	Memory effectiveness metrics	—	5.2k
127	Behavioral evaluation	—	9.3k
128	Performance tracking	—	8.8k
129	RAG-Based Agentic Memory in Code	—	32.1k
130	Procedural Memory for RAG with LangMem	—	15.7k
131	Step 1 – setting up the foundation with imports and a baseline agent	—	6.6k
132	Step 2 – define procedure structure for hierarchical learning	—	8.0k
133	Step 3 – initialize hierarchical procedural memory	—	5.8k
134	Step 4 – demonstrate learning from interactions	—	6.0k
135	Step 5 – add strategy retrieval and performance feedback	—	20.4k
136	Architectural implications and production readiness	—	5.7k
137	Advanced RAG with Complete Memory Integration	—	5.2k
138	Step 1 – create the complete agent with all memory types	—	8.3k
139	Step 2 – load synthetic investment data	—	8.0k
140	Step 3 – process baseline conversations for initial learning	—	7.9k
141	Step 4 – test improved performance and trigger adaptations	—	6.6k
142	Step 5 – complete learning progression and hierarchical retrieval	—	8.5k
143	prompt_memory: efficient single-pass learning	—	0.6k
144	gradient: critique and proposal separation	—	0.7k
145	metaprompt: multi-stage reflection for complex patterns	—	0.6k
146	Choosing the right algorithm	—	0.7k
147	Combining algorithms for comprehensive learning	—	5.1k
148	Single goal optimization: clarity with consequences	—	0.9k
149	Multi-goal balancing: realism with complexity	—	7.0k
150	Domain conversion framework	—	6.6k
151	Integration: four memory types working together	—	1.4k
152	Practical implications: enabling rapid innovation	—	2.6k
153	Get This Book’s PDF Version and Exclusive Extras	—	0.3k
154	Unlock Your Exclusive Benefits	—	1.6k
155	Why subscribe?	—	3.8k
156	Index	—	19.9k

我们自己的读书笔记(0 篇)

（还没有。读完某章后写进 docs/unlocking-data-genai-rag/notes/,那才是这本书对我们的产出。）

---

本页由 node scripts/book-build.mjs 生成:表格来自转码结果,散文来自书卡正文。不要手改本页。