GraphRAG Foundations and Architecture

Understanding Knowledge Graphs, Retrieval, and LLMs

von Wesley Harlan

262 Seiten, Taschenbuch

Kurzbeschreibung des Verlags

Traditional RAG is good at finding relevant text. But what happens when the question requires connecting facts across documents, counting events, tracing relationships, or understanding patterns across an entire knowledge base?GraphRAG Foundations and Architecture provides a practical, engineering-focused path from conventional retrieval-augmented generation to graph-based systems capable of reasoning across connected information.Starting from the foundations of LLMs, embeddings, vector retrieval, and knowledge graphs, this book progressively builds a complete GraphRAG system using Python, Neo4j, and the OpenAI API. Rather than treating GraphRAG as a collection of disconnected techniques, it explains the architecture behind the system and shows how each component contributes to accurate, grounded retrieval and generation.You will learn how to: - Understand where conventional vector RAG breaks down on multi-hop, aggregation, and corpus-wide questions.- Model entities, relationships, ontologies, property graphs, and graph databases.- Turn PDFs, incident reports, emails, and other unstructured documents into structured knowledge.- Design extraction, entity-resolution, deduplication, provenance, and graph-construction pipelines.- Build hybrid graph-vector retrieval using embeddings, seed entities, and multi-hop traversal.- Implement local, global, decomposed, and agentic retrieval strategies.- Transform retrieved subgraphs into focused context for grounded LLM generation.- Evaluate retrieval relevance, context coverage, faithfulness, answer quality, latency, and resource usage.- Assemble the individual techniques into a complete working GraphRAG application.- Prepare GraphRAG systems for continuous updates, caching, observability, failure handling, scaling, and production deployment.A single running system, Anchorline Logistics, ties the material together. Its contracts, customs incidents, and vendor correspondence provide realistic problems that grow with the architecture, allowing concepts to move naturally from simple graph operations to production retrieval and reasoning. The book also includes hands-on exercises and runnable examples throughout the progression.This book is designed for developers and engineers who have some Python and LLM experience but want to understand how GraphRAG systems are actually constructed. You do not need prior graph theory, linear algebra, or NLP research experience.If you want to move beyond "find similar text" toward systems that can reason across connected facts, this book gives you the foundations, architecture, retrieval strategies, and engineering practices to build them.

• ISBN: 9798193425458
• Sprache: Englisch
• Erscheinungsdatum: 18.08.2026
• Größe: 254 x 178 mm
• Verlag: Amazon Digital Services LLC - Kdp