SUSTAINABILITY · TECHNOLOGY · ASSURANCE

Willy Tai

AI you can trust: governed, reliable, explainable systems for regulated operations.

The Approach

AI that belongs in a regulated operation.

Working systems, not vibe-coded demos. These are agents that stand up to an audit.
Willy Tai · Singapore · Governed AI for regulated operations
Aerial view of Marina Bay Sands and the Singapore waterfront at sunrise
Singapore. Built for the operations regulators actually inspect (finance, healthcare, markets, sustainability), from the world's most tightly supervised financial centre.

I build governed, reliable, explainable AI across three fronts: sustainability, technology and assurance. The agents behave like dependable software, they make a person's judgment better, and they stand up to an audit. Every system on this site runs live in your browser, each one governed end-to-end.

The pattern is the trust boundary. Deterministic code computes every decision. The model only locates facts and phrases language. The consequential call belongs to a named human, and every action is written to a hash-chained log you can replay. Press the tamper test and the chain catches an altered record, then tells you exactly where it sits.

The newest piece turns the method itself into a working object. The Living RCM rates every workflow step's risk before controls, shows the controls in place, and re-rates what remains. Then an internal audit runs over the top and finds the control a dashboard would have called green.

Continue reading the approach →
HOW A GOVERNED DECISION TRAVELS Input UNTRUSTED TRUST BOUNDARY Deterministic engine COMPUTES · DECIDES · RATES Language model LOCATES + PHRASES · NEVER DECIDES · CITES OR ABSTAINS Named human CONSEQUENTIAL CALLS ONLY Replayable ledger HASH-CHAINED · AGENT + HUMAN IDS THE AUDIT RE-PERFORMS CONTROLS AGAINST THE LEDGER · NOT THE DIAGRAM
The trust boundary, as a process map. Code decides, the model phrases, a human owns the consequential call, and the ledger remembers. The audit tests the evidence, not the diagram.
Selected Work
VIEW ALL USE CASES →
The Living RCM · workflow canvas with risk heat map
ASSURANCE · THE METHOD

The Living RCM

Every workflow step shows its risk before controls, the typed controls in place, and the residual risk after. Then an IIA-style audit runs over the top, finds a failing control, and closes the finding.

Open demo →Read the case →
SQL Trust Gate · side-by-side wrong number versus abstain
TECHNOLOGY · THE REFEREE

SQL Trust Gate

A deterministic referee for English-to-SQL. It checks every query against a versioned semantic contract and abstains on forbidden joins, shown beside the confident wrong number a naive assistant ships.

Open demo →Read the case →
Talk to Your Data · natural language to governed charts
TECHNOLOGY · LIVE

Talk to Your Data

Type a question, get a chart. It resolves against a governed semantic catalog, so the right metric is pulled, provenance is shown, and an ambiguous word is clarified rather than guessed.

Open demo →Read the case →
What I Bring

Three ways I create value.

01 · CLIENT DELIVERY

I meet the client and carry the engagement.

I scope the problem in the client's language, deliver the build, and land it as a business number. Every case study on this site closes in an outcome rather than an architecture, because that is the conversation that wins the next engagement.

02 · CAPABILITY BUILDING

I stand up the internal machinery.

An AI practice needs adoption playbooks, secured governance processes, and the evidence infrastructure that defends both. This site is my reference implementation, built end to end by one person, so I know where every hour and every risk sits.

03 · LEARNING & DEVELOPMENT

I train the practice while I build it.

Eight years of learning and development work, teaching practising accountants through ISCA and supervising capstone projects at Nanyang Polytechnic, with a PhD on technology governance completing in 2027. People adopt what they understand, so the training is part of the build.

“AI should make a human's judgment better, or replace only the decisions that were never judgment to begin with. It should not pretend to be the human.”

See my philosophy and contact me →