Axon.
Aligning ideas
into a graph.

Strategy decays in the gap between meetings and execution. Notes go missing, decisions contradict, and three teams quietly build the same thing.

We needed a system that could ingest the unstructured exhaust of a company — meeting transcripts, Slack threads, memo drafts — and turn it into a queryable map of what the org has actually decided. Not summaries. Structure.

A LangGraph state machine routes each idea through a chain of specialised agents.

Each node in the graph is a single-purpose agent: extractor, classifier, deduper, conflict detector, owner-resolver, synthesiser. The state object carries provenance through the entire run, so every node in Neo4j can be traced back to its source utterance.

Neo4j is the source of truth. Everything is a node — intents, decisions, constraints, owners, deadlines.

Edges encode semantic relationships: SUPERSEDES, DEPENDS_ON, OWNED_BY, CONFLICTS_WITH. A natural-language query layer translates plain questions into Cypher, and a React console renders the result as both a table and an explorable subgraph.