About the Conference
NODES AI is Neo4j’s flagship annual online conference, bringing together AI engineers, data scientists, and graph practitioners for a full day of technical sessions on where graph technology meets generative AI. The 2026 edition runs 7.5 hours across three parallel tracks, with keynotes anchoring the opening and close.
The timing is good. Graph databases have gone from a niche infrastructure choice to something that increasingly shows up in serious AI architectures — not because of hype, but because the way graphs represent knowledge (as entities and relationships, not tables and rows) turns out to be exactly what language models need when they hallucinate or forget. Knowledge graphs give AI something to reason against. Graph memory gives agents something to remember across conversations. And graph-based retrieval gives RAG systems the connective tissue that flat vector search lacks.
This conference is where that frontier gets stress-tested: by practitioners who have built these things in production, researchers who are measuring what actually works, and engineers showing live code. Free to attend. Register at neo4j.com/nodes-ai.
Three Tracks
Knowledge Graphs & GraphRAG
- Zero to Agent: Hallucination-Free AI with AuraDBCharles Borderie — CEO, Lettria
- Fusing NLP and Graph: Conversational Agent for Enriched Financial DataAlix de Cremoux & Gabriel Laffitte — Banque de France
- Neurosymbolic Architecture in ProductionNeo4j, GDS, Node2Vec & LLM Grounding — João Cunha, Kipon
- Agent Reasoning with Graph World ModelsLucas Godfrey — Founder, Mindsynth
Graph Memory & Agents
- Beyond Recall: Agentic Personas with Persistent MemoryMohamed Fazil — AI Engineer, Unisys
- MemMachine: Agents That Learn, Memory That LastsChristian Kniep — Principal Architect, MemVerge
- Multi-Agent Shared Graph MemoryVaibhava Ravideshik — AI Engineer, Vy Labs
- Temporal Substrate: Persistent Identity for Autonomous AgentsConor O’Shea — Daimler Truck North America
- The AI Agent Memory LandscapeWilliam Lyon — Sr. Product Manager, Neo4j
Graph + AI in Production
- Tracing Agent Decisions with Graph EvalsAshok Vishwakarma — Founder & CTO, Impulsive Web
- Agentic AI with Knowledge Graphs on the JVMLuanne Misquitta — Neo4j
- Agent Interaction Graphs: Evaluating Multi-Agent SystemsVincent Koc — Futurist & Lecturer
- Graph-Based Long-Term Memory WorkshopAlessandro Negro — Chief Scientist, GraphAware
Keynotes
Exploring Context Graphs: From Data to Decisions
As AI scales into production, the challenge is not just smarter models — it is systems that can capture and use context effectively. This panel argues that context graphs are emerging as a new foundation for AI that can reason, learn from experience, and improve over time. Speakers include Emil Eifrem (Co-Founder & CEO, Neo4j), Philip Rathle (CTO, Neo4j), Animesh Koratana (CEO, PlayerZero), Lasse Andresen (CEO, IndyKite), and Jaya Gupta (Partner, Foundation Capital).
From Data to Knowledge to Action — The Graph Intelligence Platform
Every enterprise sits on mountains of data: documents, databases, APIs, logs. Yet most AI systems behave as if they have never seen any of it. Sudhir Hasbe (President & Chief Product Officer, Neo4j) explains how closing the gap between what organisations have and what their AI can actually use is the defining challenge right now — and how graph technology is the infrastructure answer.
Featured Speakers
Emil Eifrem Opening Keynote
Co-Founder & CEO — Neo4j
Emil Eifrem sketched the original idea for Neo4j on a flight in 2000 and has spent 25 years building the company into the dominant force in graph databases. Under his leadership Neo4j went from a Swedish startup to a company powering graph infrastructure at the largest organisations in the world — from fraud detection at financial institutions to knowledge graph systems at intelligence agencies.
His opening keynote this year centres on context graphs as a new architectural primitive: the idea that the bottleneck in AI is not model capability but the ability to connect model reasoning to the structured, relationship-rich knowledge that exists in every organisation. Whether that framing holds up under scrutiny is exactly the kind of question the rest of the conference spends the day testing.
Sudhir Hasbe Closing Keynote
President & Chief Product Officer — Neo4j
Sudhir Hasbe came to Neo4j from Google Cloud, where he led product for databases and AI infrastructure at scale. At Neo4j he is responsible for the product direction of what the company is now calling the Graph Intelligence Platform — the argument that graph technology is not just a database category but a full-stack approach to making AI systems that actually know what your organisation knows.
His closing keynote frames the same problem Emil opens with, but from the product layer: how do you go from having data to having an AI that can act on it? The graph layer between raw data and LLM reasoning is his answer, and he makes the case with production evidence, not speculation.
Alessandro Negro
Chief Scientist — GraphAware Ltd
Alessandro Negro is one of the more interesting people working at the intersection of graphs and AI. As Chief Scientist at GraphAware, he has spent years thinking about what it actually means for a system to remember — not just cache, but learn and adapt through stored relational structure. His book on GraphRAG is one of the more referenced practical texts on the subject.
Before NODES AI, he led one of the Road to NODES pre-conference workshops on graph-based long-term memory: how agentic workflows adapt through experience by maintaining and updating a graph of what they have encountered. At the conference itself he brings production examples and architecture patterns for teams building agents that need persistence beyond a single session window.
William Lyon
Senior Product Manager — Neo4j
William Lyon is one of Neo4j’s most visible technical voices, known for developer education, graph + AI tooling, and making obscure graph concepts accessible without losing rigour. His session “The AI Agent Memory Landscape” surveys where we actually are with agent memory in 2026: what approaches exist, which are production-ready, how graph memory compares to vector stores and procedural memory approaches, and what the unsolved problems look like.
If you are building agents and have not yet made a deliberate decision about memory architecture, this session is a structured way to understand what you are missing — and why the decision matters more than it might seem.
Mohamed Fazil
AI Engineer, Office of the CTO — Unisys
Mohamed Fazil works in one of the more interesting vantage points in enterprise AI: the Office of the CTO at a major IT services company, where the job is to figure out what actually works before it becomes company-wide strategy. His session “Beyond Recall: Creating Agentic Personas with Persistent, Evolving Memory” goes after a specific problem that most agent builders eventually hit — the agent that has no sense of who it is talking to, forgets last week’s context, and behaves differently in every session.
His approach builds agent personas backed by a graph memory that grows and updates with each interaction: not just storing facts, but maintaining relationships between user preferences, past decisions, and context that changes over time. It is a practical path toward agents that feel like they actually know you.
What These Topics Are About
Knowledge Graphs
Most databases store data in rows and tables. A knowledge graph stores it as a network: nodes (entities like people, companies, products, concepts) connected by typed relationships (“works at”, “part of”, “caused by”). This structure is much closer to how knowledge actually works — everything is connected to something else, and the meaning of a fact often depends on what it is connected to.
For AI systems, knowledge graphs solve a particular problem: language models know a lot about general language patterns, but they do not know your organisation’s specific data — your products, your customers, your internal processes. A knowledge graph gives the model something grounded and relationship-aware to reason against, rather than relying on what it memorised during training.
GraphRAG — Retrieval-Augmented Generation with Graphs
RAG (Retrieval-Augmented Generation) is the technique of finding relevant documents or data chunks before sending them to a language model, so the model has actual information to work with rather than guessing. Standard RAG uses vector similarity: find text that is semantically similar to the question and include it in the prompt.
GraphRAG extends this by traversing relationships. Instead of just finding similar text, you follow graph edges: if the question mentions a company, you can pull in its subsidiaries, its executives, its recent filings — all connected in the graph. This gives richer, more structured context, reduces hallucination, and makes the reasoning traceable. The trade-off is that building and maintaining the graph is work — but for knowledge-intensive applications, the quality improvement is substantial.
Graph Memory for AI Agents
When you talk to a language model, it has no memory of you by default. Each conversation starts from zero. For simple assistants this is fine. For agents — systems that take actions, run over multiple sessions, and build understanding of your context over time — it is a fundamental limitation.
Graph memory addresses this by storing not just facts but the relationships between them. An agent remembers that you mentioned liking a particular framework, that you were working on a specific project, that you changed your mind about an approach last Tuesday — all as nodes and edges in a graph that persists across sessions. The agent can then traverse this graph when deciding how to respond. The difference between a stateless assistant and one backed by graph memory is roughly the difference between talking to someone who has never met you before and talking to a colleague who has worked with you for months.
Context Engineering
Context engineering is the practice of giving language models exactly the right information, at the right level of structure, at the right moment — rather than just dumping everything into a prompt and hoping. It has emerged as one of the most consequential engineering disciplines in applied AI, because model capability is only as useful as the context you give it to work with.
In graph systems, context engineering often means deciding which subgraph to retrieve, how many hops to traverse, what relationships to include, and how to serialise that graph structure into a prompt that the model can actually reason over. Poor context engineering produces expensive, hallucinating, incoherent AI systems. Good context engineering is frequently what separates prototypes that impress in demos from products that work reliably in production.
Attend
Organiser
Neo4j
The company behind the Neo4j graph database — the most widely deployed graph database in the world. Founded in Sweden in 2007, Neo4j now powers graph infrastructure at enterprise scale across financial services, healthcare, government, and technology. NODES AI is their annual developer conference, running since 2020.
neo4j.comGraph Database Belgium
The Belgian community chapter of the global Neo4j meetup network, hosting regular events on graph databases and graph-powered applications. Based in Belgium, the group bridges Neo4j’s global conference with a local practitioner community across Brussels, Ghent, and Leuven.
meetup.com/graphdb-belgium