Build it, then measure it

I'm an AI Solutions Analyst in Columbus, Ohio. At a leadership-development institute I built the AI function from scratch: an email assistant for the CEO, a weekly operations brief, a newsletter engine and a guest-screening pipeline, all in production, and an assistant colleagues use to answer questions from ten years of the podcast.

Most of my work starts with a person doing something by hand every week: pulling numbers, reviewing copy, screening applications. I work out what they check, what they'd never let slide, and when they'd want to be asked. Then I build an agent that follows the same rules, and I measure its accuracy against answers written in advance before anyone relies on it.

The measuring runs on my own evaluation harness, a personal project I keep iterating on and use regularly. On 500-record answer keys it found a 27% error rate that a 24-record test had hidden, and measured the fix at 8.6% in its worst run. My newer work is in code, built with Claude Code: the archive assistant is a tested Python service on Google Cloud Run, and a finance operations platform (Python, FastAPI, React) spreads statements, reconciles accounts and matches invoices under one set of controls. In every build the model proposes and code decides what is safe to act on.

I’m also the person colleagues come to with AI questions. I turn the questions that keep coming back into prompt templates, guides and SOPs, so the team can work without me.

My degree is in finance (B.S.B.A., The Ohio State University, 2025), so I frame automation in terms of business impact: hours saved, errors caught, cost per item.