About
I lead anti-fraud at KOMOJU, a cross-border payments platform in Tokyo: the rule engines and machine-learning models, written in Python, that decide which merchants we onboard and which transactions we trust, and the team that builds and runs them. That covers real-time velocity and card-testing defenses with thresholds tuned by backtesting, adaptive 3-D Secure for high-risk card payments, AML transaction monitoring and an LLM agent that screens merchants. I also designed a credit-risk scorecard for rolling reserves; it stays out of production until it has been validated.
I optimize for one number: more approvals with losses held flat — fraud on payments, credit on merchants.
I want the same rigor on both sides of a risk decision: the model that makes the decision, and the system that executes it at scale.
Before KOMOJU I was CTO of a US insurance-software company and a senior engineer at Ant Group. I hold a First-class BSc (Honours) from The Open University (2025), with Distinction in all six mathematics and statistics modules, and am studying for an MSc Open there, in finance modules. I am a Node.js core collaborator (emeritus), have spoken at RailsConf and RubyConf, and am fluent in English, Chinese and Japanese (JLPT N1, 2021).
Talks and writing
- FinTech Credit and Entrepreneurial Growth
- Nijkamp et al. 2019: On the Anatomy of MCMC-Based Maximum Likelihood Learning of Energy-Based Models
- Notes from AAAI 2020 and on five machine-learning papers
- Packing your Ruby application into a single executable
- How we scaled GitLab for a 30k-employee company
- Why environment variables can make Ruby start ten times faster