The Work
Agents, automations, and the data plumbing underneath them. Everything here is real: built for the people who use it, most of it running in production today. Screenshots are redacted where client data appears.
Sales teams were reviewing call recordings by hand. This agent transcribes every call, rates the lead using multi-model routing that picks the right model for each step, and hands the team a rating with suggested next steps. Human review stays in the loop.
Runs in production on tens of thousands of leads.
Paid-media QA meant hours of manual cross-checking every batch. This tool reconciles Google Ads, Fluency, and Fabric data, flags campaign errors before anything reaches a client, and tracks error rates over time so releases are decided by data instead of gut feel.
Replaced roughly 400 hours of manual review every month.
Started as my intern prototype. Now it generates structured marketing content for national brands, with output shaped to whatever the downstream system needs: JSON, HTML, or structured data. The scale it reached made client offerings possible that were not feasible by hand.
Roughly 20,000 pages a month. It opened new client product lines.
The data lived in a system with no documented API. I watched the traffic, reverse-engineered the API, and built an MCP server from scratch so agents could reach that data through tool-calling. No vendor support, no docs, no waiting on an engineering team.
Connected AI tools to data they could not reach before.
Rebuilt the company's internal IT to be automated and cohesive. User provisioning happens automatically, the M365, Exchange, and Azure stack runs on one set of standards, and company data is structured so agents can actually collect and use it. The goal was an AI-native company, not AI tools bolted onto old plumbing.
A 65+ person company runs on it daily.
Contract work for a client. The agent researches each prospect, drafts personalized outreach, and stays customizable per campaign, so the client tunes tone and targeting without touching code. Research to send, without the manual middle.
Personalized outreach at a scale one person could never write by hand.
For the same client, I integrated their workforce-management platform (Deputy) through its API. Reports that used to be assembled by hand now build themselves at scale, and the routine employee-management work runs on automations.
Reporting at scale from a system nobody wanted to touch manually.
A navigation app for roadtrippers that trades the fastest route for the better one: scenic drives and interesting stops along the way. Built for the kind of trip where the drive is the point. In progress, and it is teaching me mobile the same way marketing taught me automation.
In progress. The fun kind of side project.
Designed and built with AI orchestration, the same way I build everything: spec, generate, review, validate. Hand-rolled canvas animation, no frameworks, two pages, hosted on my own domain.
You are looking at it.