Copenhagen Graph Databases Meetup

In this Neo4j Live session, we explore why Graph AI succeeds where LLMs fail. You’ll learn how knowledge graph–powered systems deliver significantly higher accuracy by explicitly modeling relationships, enabling traceable reasoning, and grounding AI in structured intelligence. This is the difference between fluent language and reliable logic - critical for use cases like regulatory compliance, financial analysis, and operational decision-making. If you’re thinking beyond text generation and toward AI that can reason, connect, and scale with confidence, this session will show you the path forward.

Guest: Matthias Buchhorn-Roth
LinkedIn: https://www.linkedin.com/in/mbuchhorn/
Github: https://github.com/ma3u

This workshop is designed to introduce you to Generative AI, Retrieval-Augmented Generation (RAG), and GraphRAG, and show how these techniques can be combined with Neo4j to build intelligent, graph-powered applications. You will explore how large language models (LLMs) can be grounded in both structured and unstructured data to produce more accurate, relevant, and explainable responses.
The workshop includes hands-on exercises where you will build a knowledge graph from unstructured documents and structured data, enrich it with embeddings, and use vector indexes in Neo4j to perform similarity search. You will create multiple retriever patterns—including vector-based, vector plus Cypher, and text-to-Cypher retrievers—and see how GraphRAG improves retrieval quality by leveraging graph structure.
By the end of the session, you will build a conversational agent using Neo4j, Python, and LangChain. You will leave with practical experience designing GraphRAG pipelines and a solid foundation for applying these techniques to real-world use cases.

You will learn:

• The fundamentals of Generative AI and large language models (LLMs)
• What RAG is and why it is essential for grounding LLMs
• How GraphRAG improves retrieval and response quality
• How to build knowledge graphs from unstructured documents and structured data
• How to use vectors and embeddings in Neo4j for similarity search
• How to implement different retriever patterns using the neo4j-graphrag Python package
• How to build a conversational agent with Neo4j, Python, and LangChain

This hands-on lab is designed to demonstrate how to build LLM-powered agents grounded in a Neo4j knowledge graph, without relying on traditional retrieval tools or custom orchestration code. Instead of defining tools and debugging when (or if) they are invoked, you will configure a Neo4j Context Provider that injects graph context automatically before every LLM call.
The workshop walks through three progressive configurations. You will start with full-text search using a Neo4j index to provide immediate, structured context to the agent. Next, you will extend the same pattern to vector search with embeddings for semantic matching. Finally, you will implement graph-enriched retrieval queries that traverse relationships and return connected context that goes beyond what indexes alone can provide.
Through guided exercises, you will build a working self-grounding agent that consistently receives relevant knowledge from a graph on every interaction. You will leave with practical configuration patterns you can adapt to any Neo4j schema and apply to real-world agent architectures.
You will learn:

• How to configure the Neo4j Context Provider for fulltext and vector search
• How to write retrieval queries that traverse graph relationships
• How to tune key parameters such as top_k, message_history_count, and index selection
• How to debug and validate context injection by inspecting what the agent receives

This session is designed to demonstrate how GraphRAG architectures can be applied to healthcare and other regulated domains where data is heterogeneous, distributed, and subject to stringent governance requirements. You will explore how knowledge graphs, graph neural networks (GNNs), and LLM-based agents collaborate to overcome the limitations of traditional text-centric RAG approaches, particularly in terms of temporal reasoning, explainability, and compliance.
The workshop introduces an agent-driven GraphRAG architecture built on Neo4j that unifies medical ontologies, clinical documents, and research publications in a materialized knowledge graph, while keeping sensitive electronic health record (EHR) data virtually integrated rather than centralized. Through architectural walkthroughs and examples, you will see how agents plan and orchestrate graph queries, apply GNN-based relevance and similarity scoring, and enforce controlled access to regulated data sources.
By grounding retrieval and reasoning in graph structure, this approach enables multi-hop, temporally aware reasoning with traceable and explainable outputs. You will leave with a clear understanding of when and how to combine knowledge graphs, GNNs, and agents to design GraphRAG systems suitable for high-trust environments such as healthcare.
You will learn:

• How to design an agent-driven GraphRAG architecture for regulated medical domains
• How knowledge graphs unify ontologies, documents, and reports while preserving data privacy
• How GNNs support relevance, similarity scoring, and multi-hop reasoning
• How to separate structural knowledge from patient-level data for governance and compliance
• How graph-powered RAG enables explainable, temporally grounded answers