Head of Decision Science, Asia Pacific
Live AI products, interactive visualizations, and published research.
An AI-powered equity research platform generating deep market scans with strategy classifications, analyst upside, and multi-factor scoring—fully automated.
Visit tenfold-ai.com →
Every yield instrument in Singapore in one table. Compare 37 instruments across 5 categories with yield ranges, setup complexity, and risk profiles.
Visit yieldcentralsg.tenfold-ai.com →
A walkthrough simulator for Singapore’s Integrated Shield Plan. Compare all four plan tiers with real-time bill breakdowns and hospital access maps.
Visit healthshieldsg.tenfold-ai.com →
Civilizations rendered as rivers that converge, diverge, and reshape the map of human progress across time and geography.
Explore →
Predicting World Cup Russia with ML models and tournament simulations. Full interactive visualization with real-time probability updates.
Explore →
Simulating every possible outcome to predict the winner. Featured by The Economist’s Daily Chart.
Explore →
The estimation methodology behind the predictions—logistic regression, naive Bayes, and probability modeling.
View →
Modeling international soccer match outcomes with logistic regression and exploratory data analysis.
View →Three continents, multiple industries—from banking systems in São Paulo to ML models on Wall Street to decision science across Asia Pacific. I’ve spent my career at the intersection of data and the decisions organizations can’t afford to get wrong.
Right now I’m focused on what comes after traditional AI: systems that maintain persistent knowledge, reason about scenarios, and help organizations think better. Patent holder, MIT-credentialed, and the person whose World Cup prediction model ended up in The Economist.
The hard problem isn’t storing what an organization knows — it’s encoding how it thinks. Procedures, reasoning heuristics, exception patterns, the implicit logic a senior expert applies but never documents. Current approaches flatten this into vectors and lose the structure that makes it useful.
Agents that don’t just execute — they watch themselves work, detect where their reasoning broke down, and rewrite their own cognitive architecture. Not prompt tuning. Genuine metacognition: AI that develops judgment about when to apply which approach, and when to stop and say ‘I need help.’
AI treats knowledge as static. Enterprise reality: facts expire, policies shift, context changes quarterly. The unsolved problem — systems that track when they learned something, model how confidence degrades over time, and actively distrust their own stale beliefs before they cause harm.
LLMs trained on the public internet produce generic thinking. The real edge is capturing reasoning that exists nowhere in writing — the heuristics a 20-year veteran applies intuitively, the decision patterns embedded in team dynamics, the strategic logic an organization has never articulated but acts on daily.
We can measure whether AI classifies correctly. We cannot measure whether it thinks well. For strategic reasoning, insight generation, decision support — there are no reliable metrics. If you can’t evaluate whether your AI system is actually helping on hard problems, everything else is theater.
Single agents hit walls. But the harder problem isn’t coordination — it’s trust negotiation. When specialist agents disagree, who arbitrates? When one agent’s output is another’s input, who validates the handoff? Building reliable multi-agent systems requires solving the same organizational trust problems that make human teams hard.