About Stephen Chin
Stephen Chin is VP of Developer Relations at Neo4j, the graph database company, and a long-time conference speaker and author in the Java and developer tooling world. Thirty years ago he was a Java developer migrating COBOL applications off mainframes, work he describes as multi-year projects that could never be tested thoroughly enough and often failed.
At Neo4j he speaks about context engineering and GraphRAG: using knowledge graphs rather than vector similarity alone to give AI agents accurate, multi-hop context. In an AI Engineer talk with Blitzy's Neeraj Deshmukh, he explains why bigger models and bigger windows haven't fixed enterprise code understanding, and how Blitzy's graph-backed agents mapped a 35-million-line mainframe in four days.
Notable Positions
On Context Windows
Models advertise million-token windows but deliver 200K or less of usable context, and accuracy drops as the window fills. Vector databases give similarity, relational databases over MCP give single hops; only a graph gives multi-hop, accurate context. See Context Engineering and Knowledge Graph.
On Legacy Modernization
Code is a graph-shaped problem. Once the estate is a graph with business rules and processes on top, the same layer supports DORA compliance, business rule tracking and fraud use cases beyond the original migration.
Key Quotes
- "Context is serial, but information has relationships." (on why graphs fit agent context)
- "If you make one little line of code change, you risk breaking the whole system." (on mainframe estates)
Related Reading
- Knowledge Graph - The structure Chin advocates for agent context
- Context Engineering - The discipline his talks address