See the organisational mess
Search synthetic emails, Teams-style messages, operational data and policies. Trace conflicting claims back to specific records, departments and reporting periods.
Prototype 04 / Enterprise AI experiment
The Meaning Layer.
Before AI can answer, can it understand?
Imagine a 5,000-person company with competing definitions, fragmented systems and decades of accumulated knowledge. Connect an AI to everything and you'll get more information. But will you get the right meaning?
● Fictional company and synthetic evidence · Free to explore · No sign-in needed
The problem isn't simply data. It's meaning.
One team calls production the tonnes mined; another means saleable tonnes dispatched. Finance describes actual cost one way and Procurement measures committed spend another. A general-purpose model can read these documents, but without context and governance, it can still misinterpret the business.
Work out what language means
Explore candidate business concepts, aliases, owners and connections. Try questions that cross functional boundaries and see where simple keyword matching breaks down.
Decide what's trustworthy
Separate genuine contradictions from different measurement scopes. Experiment with human validation before interpreting evidence for an executive decision.
From messy sources to accountable answers.
DiscoverMap sources
IngestRespect access
InterpretInfer candidates
ReconcileResolve conflicts
ApproveHuman governance
ReasonCited decisions
The experiment explores more than document search. It asks whether organisations can establish a shared, versioned meaning layer that makes AI more reliable without ignoring local business context.
Try Redgum Resources.
Open the lab and ask: “Why is production below plan at Pilbara North?” Then investigate the source evidence, semantic differences, graph relationships and steward decisions yourself.
What is real today? The source explorer, terminology linking, deterministic search, graph projection and certain conflict checks run against a generated synthetic dataset. The executive narratives are curated; model-assisted discovery and Q&A are being developed and will be introduced only with strict usage controls.
This is a research and design prototype, not a production enterprise AI system. It uses fictional people and records, does not access workplace data and makes no claims about a real mining operator. Future live model capability will be clearly identified as experimental.