Giving LLMs a Map: Building Smarter GenAI with GraphRAG

Jennifer Reif, Developer Advocate, Neo4j TALK DESCRIPTION: Generative AI is powerful, but without the right data and retrieval strategies, results can quickly break down. This session will explore how GraphRAG combines knowledge graphs with retrieval-augmented generation to deliver more accurate, context-rich AI applications. Through live demos and code, we will walk through building a GenAI solution end to end using Neo4j and Python. Learn how to construct knowledge graphs from unstructured and structured data, make key design decisions around schema and chunking, and implement multiple retrieval strategies—including vector search, vector plus Cypher, and text-to-Cypher approaches. Then, pull all these skills together in a conversational agent built with Neo4j and LangChain. Come to this session and leave with practical techniques for designing knowledge graphs, choosing the right retriever for a use case, and applying GraphRAG patterns you can adapt to your own GenAI projects. A good starting point for code to be discussed is in the examples in: https://github.com/neo4j/neo4j-graphr... SPEAKER BIO: (presenting remotely) Jennifer Reif is a Developer Advocate at Neo4j, speaker, and blogger with an MS in CMIS. An avid developer and problem-solver, she has worked with many businesses and projects to organize and make sense of widespread data assets and leverage them for maximum business value. She has expertise in a variety of commercial and open source tools, and she enjoys learning new technologies, sometimes on a daily basis! Her passion is finding ways to organize chaos and deliver software more effectively.   / jmhreif   https://www.meetup.com/sf-bay-acm/eve... In-person, Zoom and YouTube. Speaker will be presenting remotely, SFbayACM will support a local audience at VRP in Mountain View, CA 0:00 Chapter Intro 8:09 Speaker Intro 9:30 Presentation: Giving LLMs a Map: Building Smarter GenAI with GraphRAG 9:55 About Jennifer 11:29 The Promise of GenAI 19:56 GraphRAG Process: 19:59 Step 1 Prepare data source 56:09 Step 2: Plan retrieval 1:26:14 One step further - Agents! 1:33:38 Key Takeaways 1:34:42 Resources 1:36:37 Q&A AI Summary This video, presented by Jennifer Reif, discusses the concepts, techniques, and tools behind GraphRAG (Graph Retrieval-Augmented Generation), specifically using Neo4j and Python to build smarter Generative AI applications (9:30-10:00). Video Highlights and Core Concepts: The Promise of GenAI: While powerful, LLMs require accurate data and context, which traditional RAG (using vector search alone) may struggle to provide when relationships between data points are crucial (11:29-19:56). The GraphRAG Process: Data Preparation: The video outlines steps for preparing data sources and constructing a knowledge graph (19:59-56:09). Retrieval Planning: Jennifer demonstrates multiple retrieval strategies, including vector search, vector plus Cypher queries, and text-to-Cypher approaches to retrieve precise context (56:09-1:26:14). Building Agents: The presentation moves beyond simple retrieval to show how agents can handle decision-making, using tools to query the database and summarize results for complex questions (1:26:14-1:33:38). Key Takeaways: Data design matters significantly for GenAI system performance. Knowledge graphs provide context that semantic vector search alone cannot. Different types of questions require different retrieval strategies, often combined within an agentic framework (1:33:38-1:34:42). Resources: Code examples are available on GitHub (https://github.com/neo4j/neo4j-graphr.... Additional learning is available via Neo4j GraphAcademy and the Nodes AI virtual event (1:34:42-1:36:37).