AI Risk & Governance
Translating frameworks — MAS guidelines, the EU AI Act, the NIST AI RMF, IMDA's Agentic AI guidance — into controls teams can actually apply.
AI Risk · Applied AI · Data Science · Business Intelligence · Singapore
From building AI to governing it.
Nearly a decade building AI across industry and research.
Now I'm focused on how we bring AI into the world safely and ethically — and on making AI risk and governance something you can explore.
01 I came to AI governance by building AI .
02 You can't govern what you don't understand .
03 So I make how it works — and breaks — easy to explore .
About
These days, my focus is AI risk in regulated finance.
I came to it as a builder and researcher.
For nearly a decade I built AI across industry and academia, spanning e-commerce, research, an early-stage Web3 startup, and now regulated finance. What stayed constant was a focus on practical systems built with a business mind, using machine learning, optimisation, and business intelligence to turn data into decisions.
Moving into AI governance didn't mean leaving engineering behind. For me it is the same discipline applied to risk. Because I have built these systems myself, I know how they are made, where they fail, and what responsible AI has to mean once a product is in production.
I treat AI governance, risk, and safety as a field I am still actively learning. To make sense of it, I build open, interactive tools that turn these ideas into something easier to grasp. Some map how AI systems, risks, and controls connect. Others are study companions that make dense frameworks easier to navigate.
- PhD, Computer Science — Nanyang Technological University (Alibaba–NTU Joint Research Institute)
- M.Tech, Artificial Intelligence — National University of Singapore (Valedictorian)
- Doctoral research: sample-efficient deep reinforcement learning via generative models
What I focus on
Where my attention goes now.
Agentic AI Safety
Mapping how agentic systems fail and how to defend them: attack and defense flows, guardrails, and the failure modes behind OWASP LLM Top 10 and MITRE ATLAS.
Governance you can use
Building interactive tools — explorable atlases, RACI accountability maps — that turn dense policy into something practitioners can navigate.
Research
Research I've published.
First author on four peer-reviewed papers in reinforcement learning and applied machine learning. The reinforcement-learning line — from my PhD at Nanyang Technological University — was co-authored with Frans A. Oliehoek (TU Delft), Athirai A. Irissappane, and Jie Zhang (NTU). Full profile on Google Scholar
- 2023
Teacher-Apprentices RL (TARL): Leveraging Complex Policy Distribution through Generative Adversarial Hypernetwork in Reinforcement Learning
Autonomous Agents and Multi-Agent Systems (JAAMAS), 37(2):25 Journal
- 2021
Learning Complex Policy Distribution with CEM Guided Adversarial Hypernetwork
AAMAS 2021, pp. 1308–1316 Full paper · 25% acceptance (152/612)
- 2021
Improving Sample-Based Reinforcement Learning through Complex Non-Parametric Distributions
AAMAS 2021, pp. 1837–1839 Extended abstract
- 2019
Development of Wearable Gait Assistive Device Using Recurrent Neural Network
IEEE/SICE Int. Symposium on System Integration (SII) Most cited · 21 citations
Selected work
Things I've built.
Risk AI RiskAtlas
AI systems, risks & controls
An interactive learning lab for how GenAI and agentic AI systems work, where they fail, and how to defend them. Explorable system diagrams and step-by-step attack/defense simulations, with a risk taxonomy and control library mapped to OWASP LLM Top 10, MITRE ATLAS and the NIST AI RMF.
Governance IMDA Agentic AI Governance Study
A framework, made navigable
An interactive study companion that turns Singapore IMDA's dense Model AI Governance Framework for Agentic AI (v1.5) into navigable topics and a calibratable RACI accountability matrix — every claim traceable to source, with a worked financial-institution example.