​Automatic Ontologies: Build the Context Graph for Trustworthy AI Agents, Yann Bilien, Rippletide

In this part of the talk, the speaker explains that building context graphs was actually a huge pain point at first. Their team spent 18 months in R&D, working with some bright minds and testing real enterprise deployments, to figure out how to automate graph creation—especially using ontologies. They share real production examples, like a testing agent that checks code every time someone commits to a repo, larger data-heavy graphs, and even a credit decision system for a car manufacturer. What makes these context graphs different is that they combine three things: structured knowledge (entities and relationships), built-in processes so decisions aren’t left entirely to language models (avoiding issues like hallucinations), and guardrails that enforce clear, deterministic rules. Even though a lot of people are interested in using context graphs, far fewer have actually deployed them in production—mainly because setting them up is tough, especially when integrating multiple data sources and managing large, complex graphs.