Work
Products, and the calls behind them.
Each of these is written as a decision, not a résumé line: what the problem actually was, what I chose to do about it, and what happened next. Numbers are at the top.
Case studies
06 totalEmployee Works: zero to general availability
A brand new product surface taken from concept to GA in six months, then scaled to ten enterprise pilots and more than a million end users.
People Finder AI Agent
Chose a knowledge graph over vector RAG for organisational search, because "who knows about this" is a relationship question, not a similarity question.
Notifications and Feedback
Owned the notification platform behind hundreds of millions of monthly sends, and a feedback product employees chose to use rather than were told to.
My Requests
Rebuilt the request experience and its underlying architecture, collapsing close to a hundred recurring case tasks a year down to about ten.
Enterprise Policy Connector
A retrieval layer over enterprise policy documents, built to explore what governed retrieval looks like when the source material is authoritative and the answers have to be defensible.
Cross-Jurisdiction Compliance Agent
Benchmarked three agent architectures against each other on the same compliance task, to find out which one degrades most gracefully when the rules conflict.
Also built
GitHub →Currently available
Want the longer version of any of these?
I am happy to walk through the decisions in detail, including the ones that did not work.