Estate Twin
A 3D twin of a small estate. Inject a fault, watch it reach the floors that machine serves, and approve the work order.
Every demo runs on synthetic or public data, labelled as such. Rules make the decisions, the model proposes with a stated confidence, a named person approves, and every step is logged. Press Run it here to open one in place, or Open demo for the full page.
The AI demos (data room, go-live gate, vendor scorecard, talk to your data) run on the data these systems produce.
A 3D twin of a small estate. Inject a fault, watch it reach the floors that machine serves, and approve the work order.
A fault photo becomes a work order. The model proposes the fault class, rules set priority and SLA, and a named person approves.
Air-handler and rooftop-unit faults diagnosed in public test data from two US national labs, without the answer key, then scored against it.
850 synthetic work orders mined into the as-is maintenance process. The CMMS reports 93.8% SLA compliance; the clock says 73.4%.
Six ways a building is cooled, each drawn as a working schematic. Hover any pump or valve, run the start-up sequence, drag the building load.
124 synthetic devices checked against seven lifecycle rules, with a risk register, a twelve-month calendar and an onboarding gate.
1,157 synthetic building alarms in one week reduced to the 29 that need a person. Two cascade storms traced to their source.
Three source systems that disagree, profiled, quality-gated and reconciled, with a person deciding the two uncertain merges.
Five acceptance criteria fixed and hashed before any result is seen. Week-six drift breaks the signed operating point and withdraws the approval.
24 due-diligence questions where a claim without evidence scores at most a third. The sample vendor claims 77% and evidences 31%.
A twelve-requirement climate checklist where every verdict must quote the report or abstain, re-scored the way FY2029 limited assurance will.
Type a question and get a chart from a defined metric catalogue, with provenance shown and ambiguous terms sent back for clarification.
Generated SQL checked against a semantic contract before it runs. Where the model abstains is shown next to where it would have returned a confident wrong number.
Every disclosed emissions figure recomputed from activity data and versioned factors, to the gram.
About the data. Method demonstrations built for this site by one person. No client data, no client systems. Sample photographs are AI-generated and say so; the Plant Diagnostician uses public fault-injection datasets from two US national laboratories. The delivery record behind them is on the track record page.