Knowledge Design in the Age of AI: Graph Technologies in Practice (Meetup #55)

This meetup explores how graph technologies support knowledge design in the age of AI.
From vector search and graph algorithms to schema design and AI-driven knowledge extraction, the talks share real-world lessons on modeling complex, interconnected data.
The central theme:
How do we design connected, reliable knowledge — not just store data — in AI-driven systems?

Agenda:
12:45pm–1:00pm
Registration & Networking
1:00pm–5:00pm
Featured Speaker Sessions

• 🎤 Kozo Nishida (RIKEN) — Getting Started with Vector Search in Neo4j
• 🎤 Sixing Huang (Gemini Data Inc.) — TBD
• 🎤 Hiroki Kokubo (JGC Corporation) — Schema Design for Complex Data (P&ID Lessons)
• 🎤 Benjamin Squire (Audience Acuity) — Working with Graph Algorithms
• 🎤 Jason Koo (Consultancy) — Building Knowledge Graphs from Documents with AI
• 🎤 Koji Annoura (Neo4j Ninja) — Underwater Heritage and Knowledge Integrity in the Age of AI

Closing Discussion

✨ What to expect:

• Practical talks grounded in real projects
• Concrete examples using graphs, AI, and vector search
• Honest lessons from production systems
• Technical insights balanced with broader knowledge design perspectives

Ideal for engineers, data professionals, and anyone interested in building meaningful knowledge systems with AI.

Special Thanks:
Our sincere thanks to Fukuoka Growth Next for their generous venue support.

Topics/Speakers:
🎤 Kozo Nishida, Technical Scientist, RIKEN
Talk Topic: Getting Started with Vector Search in Neo4j
Talk Description: As AI-driven applications continue to advance, vector search has become an essential capability for modern databases.Unlike traditional keyword-based search, vector search retrieves data based on semantic similarity by comparing embedding vectors rather than exact text matches.Neo4j now offers vector search as a core feature, enabling developers to combine graph structures with similarity-based retrieval within a single platform.In this session, we will introduce the vector search capabilities available in Neo4j through simple datasets and real code examples. Attendees will gain a practical understanding of how vector search works in Neo4j and how it can be applied to real-world use cases.

🎤 Sixing Huang, Bioinformatician, Gemini Data Inc.
Talk Topic: Healing the Language Gap: Building a Multilingual Medical Dictionary with Neo4j
Talk Description: Medical vocabularies like SNOMED C, ICD-10, and UMLS are the backbone of healthcare interoperability worldwide — but their linguistic reach has a serious blind spot: they offers no support for Chinese queries at all, leaving the healthcare needs of over 1.4 billion people effectively invisible to global medical informatics tools.
In this talk, I will walk through how I built a multilingual medical dictionary as a graph database in Neo4j, enabling traversal of ICD-10 and SNOMED concept hierarchies in Chinese, with Spanish and German as bonuses. I'll show why a graph database is a natural fit for this problem: the hierarchical and associative relationships between medical concepts map elegantly onto nodes and edges, and Cypher makes it straightforward to query across language boundaries and climb or descend the concept tree in a single query.
You'll leave with a clear picture of how to model controlled medical vocabularies as a property graph, practical strategies for ingesting and linking multilingual terminology data, and a sense of how graph-native traversal unlocks the kind of cross-lingual clinical queries that relational databases struggle with. Whether you work in healthcare, health insurance, or are simply curious about knowledge graphs, this talk will show you what Neo4j can do when language is no longer a barrier.

🎤 Hiroki Kokubo, Engineering Data Architect, JGC Corporation
Talk Topic: Schema Design for Working with Complex Data in Neo4j - Lessons Learned from Modeling P&ID Data -
Talk Description: In this talk, I will share the trial-and-error process of designing a graph schema while introducing Neo4j into a production system. The source data comes from plant engineering diagrams (P&ID: Piping & Instrumentation Diagrams), but the core topic is a challenge common to many Neo4j users: how to translate complex, highly interconnected data into an effective graph model.Based on hands-on experience—from building the Neo4j schema to developing and operating the system on top of it—I will highlight where a mindset shift from relational thinking is necessary. I will also discuss practical design decisions such as node granularity, how to structure relationships, when to model something as properties versus relationships, and how query patterns in Cypher (traversals and aggregations) should influence schema design. In addition, I will touch on schema considerations for future use cases such as GraphRAG.Rather than presenting polished best practices, this talk focuses on real operational lessons learned, including what worked, what didn’t, and what we revised along the way. I hope it will be useful for those planning to embed Neo4j into business systems or struggling with schema design.

🎤 Benjamin Squire, Principal Data Scientist, Audience Acuity
Talk Topic: Working with Graph Algorithms: Identity Resolution
Talk Description: The talk will go over details about what I do at Audience Acuity, talking about building an Identity graph from raw sources, why we chose Snowflake powered by Neo4j for our Graph Algorithms, and the outcomes we achieved.
It will showcase our graph model, how we use Weakly Connected Components, and how we scale with parallel processing in Snowflake to build a graph with 3.8 B nodes and 12 B edges.

Ben is a Principal Data Scientist at Audience Acuity. His career started 10 years ago in Media and Ad technology with special emphasis on Identity Graphs for enterprises.

🎤 Jason Koo, Fractional DevRel, Jason Koo Consultancy
Talk Topic: A Little Structure Goes a Long Way: Building Knowledge Graphs from Documents with AI
Talk Description: If you already know how to design a graph schema by hand, you know how much thought it takes. In this session, we take a different starting point: raw PDF documents. Using Lettria, we will extract entities and relationships from multiple documents and load them into Neo4j — first with no guidance at all, then with a lightweight ontology to steer the process, to illustrate the output difference. A small amount of upfront structure improves the quality and connectedness of the resulting graph. We explore the results visually with gdotv, run cross-document queries, and discuss how this approach fits alongside traditional schema-first workflows.

🎤 Koji Annoura, Neo4j Ninja, Graph Data & AI Enthusiast
Talk Topic: From the Ocean Floor to the Future — Underwater Heritage and Knowledge Integrity in the Age of AI
Talk Description: Beneath the ocean lie shipwrecks, cargo, and traces of ancient trade routes — fragments of cultural memory that connect us to the past. Yet much of this knowledge remains scattered across reports, academic papers, and institutional archives, often unstructured and difficult to access.
Generative AI can now produce fluent explanations of history and culture. However, its outputs depend entirely on the structure and reliability of the underlying data. When information is fragmented or ambiguous, AI systems may fill in the gaps — sometimes generating narratives that sound convincing but lack factual integrity.
Using underwater archaeology as a case study, this session explores:

• Why information should be modeled as interconnected knowledge rather than isolated records
• How linking historical context, geography, people, and trade networks strengthens reliability
• Why connecting data to open knowledge ecosystems such as Wikidata matters
• Why structured public knowledge infrastructure is essential in the age of AI

This talk is not about promoting a particular technology. Instead, it shares a perspective on knowledge design — one that does not rely solely on convenient AI outputs, but focuses on building durable, verifiable, and reusable knowledge.
How can we preserve the histories resting beneath the sea not as temporary generated text, but as trustworthy knowledge that future generations can examine, reuse, and reinterpret?
Let’s explore that question together.