Managing Kubernetes at scale means asking the same questions over and over: which pods are failing, what changed in this service, and can I move workloads safely? Artavazd (Art) Balaian, a software engineer at Agoda, built Ariadne to make those questions easier to answer by turning live Kubernetes state into a Memgraph property graph that can be queried with Cypher and exposed to AI tools.

In this session, Art will walk through Ariadne’s architecture: a Rust-based ingester that snapshots cluster state and applies incremental updates into Memgraph, a graph model designed for deeply nested Kubernetes objects, and an MCP server that gives agents a safe interface over the data. He’ll also briefly cover the experimentation behind the AI layer, including evals, prompt search, and validation techniques for natural language-to-Cypher workflows.

Art will explain why Memgraph was a good fit for this problem. Kubernetes objects are irregular, deeply nested, and full of maps, lists, and annotations. Memgraph’s schemaless model and flexible property support made it possible to preserve that structure without heavy denormalization or losing queryability.

If you work in platform engineering, SRE, or graph-powered developer tooling, this talk offers a practical look at building a live Kubernetes graph system and connecting it to AI through MCP.

Building a Kubernetes Graph Engine for Agents