About

I didn’t start out in data. My first job was as a production-engineering research engineer at a car manufacturer, working on things like the ordering and precision of machining lines. One day I thought, “what would this look like if I treated it as data?” So I taught myself analysis and machine learning, proposed a project to automate and visualize line management, and saw it through to deployment on the factory floor. That’s where my interest in data started.

I carried that experience into a move to data science in finance. As a data scientist I’ve worked on credit (underwriting) score models for products like card loans, on improving review rules, on delinquency analysis for buy-now-pay-later, on targeting for card marketing, and on image matching to help curb financial crime. I also ran internal training to teach AI to colleagues.

These days I build underwriting models at a fintech group in Japan. My day-to-day is spent with credit-review data, purchase data, app-usage history, and the like.

This blog grew out of what I noticed after moving into finance data science. I had taught myself from ML books, but something never quite lined up in underwriting. It turns out this field runs on slightly different rules than data science in general, and I want to lay that difference out in the language of practice. Less “teaching” than sharing what I’ve learned on the job — and I’d genuinely welcome advice from people who have worked in this field far longer than I have.