I use production data to fix the problems users never report: a silent bug that dropped requests without any error, a latency regression hidden by automatic retries, and inconsistent behavior across servers.
How I got here
I started on the back end of a time-and-attendance product, building the rules engine that turns raw employee punches into compliant records. That job teaches you one thing early: when software touches someone's pay, the rules must be exact and every action must be explainable.
When AI agents arrived, I kept that habit. The model proposes. Clear rules, checks and permissions decide. Anything that changes a record waits for the person to confirm.
Today I'm the technical lead for an employee-facing AI assistant built on the Strands Agents framework. I took it from architecture to launch, cutting attendance recording from 3–4 seconds to 1–2 seconds end to end, with confirmation before anything is saved.
- 2018Built the rules engine and REST services that process employee time data.In AI at the time: Foundation models. Knows a lot, can't do much on its own.
- 2021Senior developer: performance work, feature ownership, mentoring and design reviews.In AI at the time: Generative AI. Fluent and creative, but it can make things up.
- 2023Lead: moving the team from services to production AI agents.In AI at the time: Retrieval (RAG). Answers grounded in real documents.
- NowTechnical lead for an AI assistant that acts on users' behalf.In AI at the time: Agents. Use tools, follow rules, and ask before acting.