I turn messy data into decisions people make: models judged by what they change, and AI that drafts while code checks the numbers.
Every project splits the job the same way. Pick a role to see who does what, or open a project for the full story.
A fraud model judged by the fraud it stops, not by test accuracy. Validated on months it never saw, then shipped as a planner and an analyst queue.
Cited answers for financial analysts. Every figure is checked against its source, and every source against who is asking.
AI quoting for small caterers. The assistant handles the conversation, tested code handles the price, and the owner approves every quote.
Pick a message and watch who handles it. These are real behaviours from my projects.
Hover a tool to see where I've used it.
I like ambiguous problems with a real decision at the end of them.
My rule of thumb: let AI handle language, let code own the numbers, and let a person make the call. It's how a team ends up trusting a tool, and actually using it.
Built AI-powered planning and analytics tools for sales teams and took them from prototype to production.
Integrated LLM-driven sourcing with SQL and Python, cutting screening time 25% and surfacing 40+ acquisition targets.
Data and business analysis for client engagements.
Built CLV and churn models for high-net-worth segments and automated the pipelines behind retention strategy and revenue planning.
Machine learning, data engineering and AI, with most of the projects on this page built along the way.