Meet the team: William Lyon

In a recent GraphGeeks interview, Amy Hodler sat down with Hypermode's Will Lyon to discuss his fascinating journey through the graph database space, from his early days with Neo4j to his current role at Hypermode working with Dgraph. The conversation reveals insights into the evolution of graph databases and their emerging role in model-native apps.

Watch a video of their interview in the video below – or read on for lightly edited transcript of their discussion:

From Neo4j to Dgraph

Will's graph database journey began in 2011 when he stumbled across Neo4j at a hackathon. While pursuing his master's degree, he and his team built a GitHub repository recommender system using Neo4j, which later became the foundation for his thesis on link prediction using triadic closeness metrics. His involvement deepened through a Google Summer of Code project, where he worked on extending Neo4j's geospatial capabilities.

After eight years at Neo4j, Will's path led him to Dgraph, which takes a fundamentally different approach to graph databases. While Neo4j began as an embedded Java database optimized for local graph traversals and later added high availability and horizontal scaling capabilities on top of its existing architecture, Dgraph prioritized scalability through distribution from its original design, heavily influenced by Google's highly scalable Spanner database. The "D" in Dgraph stands for "distributed," reflecting its architecture that enables horizontal scaling and automatic sharding using a purpose-built, distributed key-value store called Badger.

Will shares that one of the most interesting aspects of Dgraph is its hybrid approach to graph models. While it exposes a property graph model to developers at query time, it internally works with triples (subject-predicate-object relationships). This design choice allows Dgraph to optimize for large-scale traversals while maintaining familiar property graph concepts for developers.

The rise of AI and knowledge graphs

At Hypermode, our focus is on model native apps - applications that treat AI models as first-class citizens. Through our open source framework, Modus, we're working to integrate knowledge graphs with AI workflows, particularly in three key areas:

• Basic RAG (Retrieval Augmented Generation) pipelines
• Tool use and function calling for AI-driven control flow
• Knowledge graph RAG for enhanced context in AI applications

From Will's perspective, the applications for these technologies has evolved rapidly in the last couple years. Initial use cases focused on enhancing search relevance and recommendations in existing applications. However, there's now growing interest in using graph databases and AI for:

• Financial services and fraud detection
• Workflow automation with human oversight
• Creative processes requiring both AI assistance and human expertise

"I feel this renewed interest in graphs in the context of knowledge graphs for AI workflows... going forward, developers will use models in their apps. This concept of a model-native app is where you have abstractions in the development frameworks that you're using that treat models as a first-class citizen."

Looking forward

From his own experience, Will sees how the graph database landscape continues to evolve, particularly as AI applications become more prevalent. Rather than replacing human expertise, the focus is increasingly on building systems that enhance human capabilities through the combination of knowledge graphs and AI.

Will's journey reflects the broader evolution of graph databases - from specialized tools for specific use cases to fundamental components of modern AI-enhanced applications. As he notes, the graph community remains tightly connected, with relationships and collaborations spanning years and companies, creating unexpected connections that continue to shape the industry's future.