A community that thinks — but never guesses
A ledger of real events is the perfect substrate for useful intelligence — and the perfect temptation to let a model start making things up. Here is what AnKampung reaches for, honestly labelled, and the three refusals that keep every one of them subordinate to the truth.
Three refusals, before any model
A platform for homes, money and a gate must earn trust by what it will not do. These come first, and every capability below is built to obey them.
Eight ideas, honestly labelled
Live in pilot runs today; in build is engineered against the real ledger; concept is a direction named early — and all of them sit under the engine that never invents facts.
The engine that never invents facts
Not a model — the foundation every 'smart' feature is allowed to sit on and never override. State moves only along legal transitions; a contradictory event is rejected or quarantined with a reason. An AI suggestion can propose; it can never write a fact the ledger can't prove.
Ledger anomaly detection
Because every change is an immutable event, the system can watch its own books: a re-delivered payment that would double-credit, an arrears figure that doesn't reconcile to the receipts, a state that jumped illegally. It flags for a human before an audit would ever find it.
Predictive maintenance triage
Work orders carry a history the engine already keeps — recurring lift faults, a floor pump that fails every monsoon, an SLA quietly slipping. A model ranks and routes before the queue piles up; the committee still assigns, and the state machine still refuses to skip a step.
The offline guard copilot
At the gate, inference runs locally on the guard's device so a real visitor is never stuck behind a dropped connection. When the network returns, what happened in the dark reconciles against the spine with origin, device sequence and wall-clock preserved — helpful, and honest about what it did offline.
Visitor pattern & tailgating signals
Normal gate flow has a shape. Learning it lets the guard's app surface an anomaly — a tailgated entry, a pass used far from its window — as a prompt to look, never an auto-lock that punishes a resident whose QR is fine. Suggestions to a human, not actions on a person.
Plain-language committee reports
The month's ledger, work orders and notices, summarised into a board read a resident can actually follow — where the sinking fund stands, what got fixed, what's overdue. Generated from the twin's own proven facts, so the prose can't say more than the events support.
Trilingual circular drafting
A committee member states an intent; the tool drafts the announcement in Bahasa Malaysia, English and Chinese to match the community's screens. A human approves before send — the model drafts, the platform never auto-publishes to a whole enclave.
Procurement & demand-sense
Facility bookings, maintenance spend and utility load already pass through the ledger as facts. Aggregated honestly, they inform the annual budget and the next procurement — forecasting on real events, not on a vendor's optimistic spreadsheet.
The discipline isn't decoration — it traces to the Strata Management Act, the incumbent community apps, an advertising-model cautionary tale and the honest computer science of event sourcing. The floor is on the Evidence page.
See the evidence floor→Ready to build something intelligent?
Let's engineer your next product — from idea to launch.