SMART CITIES · AI · SUSTAINABILITY
Willy Tai
Articles

Articles on digital twins, AI governance and estate operations.

The argument behind the demos. What the standards require, where the line between code and model belongs, and what an estate owner should put in a tender.

01Series

Series · four parts · complete

The Minimum Twin

Four articles on how little digital twin an estate needs to run its monitoring loop. It starts with a twin you can operate in the browser, removes most of it, rebuilds it from one photograph, and ends with what an owner should put in a tender.

Start with Part 1
  1. 01A twin you can operate
    A live twin in the browser.
  2. 02State, kind, rough location
    The Lattice view removes most of the model and still catches the fault.
  3. 03A twin from one photograph
    The twin's geometry taken from one photograph through a table that marks every value observed, inferred or assumed.
  4. 04What to put in a digital twin tender
    Which part of a twin budget buys the twin, and which buys polygons?

02All articles

Newest first
SEP 2026
THE MINIMUM TWIN · PART 4What to put in a digital twin tender
Which part of a twin budget buys the twin, and which buys polygons? Five clauses an owner can write instead.
 AUG 2026
THE MINIMUM TWIN · PART 3A twin from one photograph
The twin's geometry taken from one photograph through a table that marks every value observed, inferred or assumed. The photograph is AI-generated and labelled.
 AUG 2026
THE MINIMUM TWIN · PART 2State, kind, rough location
The Lattice view removes most of the model and still catches the fault. On the Gemini Principles, and the three answers a monitoring twin owes you.
 AUG 2026
THE MINIMUM TWIN · PART 1A twin you can operate
A live twin in the browser. Rules detect the fault, the model proposes a diagnosis with its confidence shown, a named person approves, and a hash-chained log records it.
 JUL 2026
COMMENTARYWhat ISO 42001 certification does and does not cover
A close read of ISO 42001 and the EU AI Act. The certificate covers the management system. The evidence trail is what you keep.
 JUN 2026
FRAMEWORKA control framework for agentic AI under MAS rules
A practitioner's mapping of agentic-AI controls to MAS FEAT, Veritas and model-risk guidance, set out as objective, control, evidence, owner.
 JUN 2026
COMMENTARYThe trust boundary
Whether AI can be governed is settled in the architecture before any policy is written. On where to draw the line between what code computes and what a model may say.

More to come.