About
My team at Commonwealth Bank builds agentic systems: customer-facing agents for investment education and trading guidance, along with scam detection, entity resolution, and the safety and risk evaluation a regulated environment requires. I support these projects at every level, from introducing stakeholders to agentic AI through to deployed systems, and I do hands-on work where needed.
A customer-facing agent has to be fast and cheap enough to be worth using, and safe and trustworthy enough to be allowed near a customer. Where you land between those two is a design decision, not something you discover afterwards, and most of what goes wrong is in the details. A small language model can be expensive when it holds a whole GPU for two thousand calls a month. Adding a subagent adds latency for the whole system. Retrieval is easy for simple things and a different beast at the edge cases and in compliance.
For a decade before industry I studied human decisions with psychophysics. A person cannot tell you how they decide, so you present controlled evidence, record the choices, and infer the mechanism from the pattern of errors. The study I return to most began at RIKEN in Japan and finished at Stanford. We found that people do not weigh each new piece of evidence on its own. They carry their last few trials with them, and failure weighs differently from success. It was published in PNAS in 2016.
I came to the bank through AI Labs, its applied research team, working on problems that went through to production and were novel enough that three papers came out of them.
The question is the same in both places: when does a system have enough evidence to answer, and when should it hand the decision to a person? Psychophysics separates how well a system discriminates from how willing it is to say yes, and accuracy alone misses that difference. When a problem resists solving, I treat the assumptions behind it as hypotheses and test them.
I am based in Sydney, Australia. If you are building AI systems that help people, email is the best way to reach me.