Selected work
AI shipped into real operations
Not AI demos. Systems built and run inside real workflows, an editorial platform, a memory-care floor, a live e-commerce business, a client's own cloud. I direct AI the way a technical director runs a team: define the architecture, push back when the plan is wrong, ship what works.
Products and platforms
Working build, verified on Bedrock
StyleProof
An AI copy editor and cited fact-checker that deploys into a customer's own cloud.
Live, revenue-generating
Petropolitan
A live AI pet-portrait business, built and operated end to end, with measured quality.
In daily production use
Bloomday
An AI activity planner in daily production use on a memory-care floor.
Working build, device-verified
MacWrangler
Agentic Mac-fleet intelligence over a custom read-only MCP server, safe by construction.
Deployed, in use
Job Concierge
A deployed multi-model AI job-search platform with per-user RAG isolation.
Live, in-browser
DropCheck
In-browser file forensics that never sees a server. Nothing uploads.
Client engagements, anonymized
In production for the client
Disaster recovery, built with AI
Ransomware-immune multi-region DR for a media production company, built solo, directing AI.
Deployed to the fleet
Wi-Fi fault telemetry
Fleet-wide Mac Wi-Fi telemetry that isolated faults a dedicated network team could not.
In production
Editorial platform to AWS
Moved a 200-user editorial platform to AWS around live operations, with zero disruption.
Personal tools, built and used daily
In daily use
Spitballer + Idea Critique
A two-app pipeline that turns a half-formed idea into a spec seven adversarial AI experts have already torn apart.
Working build, in active development
Agent Maestro
A macOS command center that conducts a fleet of Claude Code sessions in one window, with governance gates on anything irreversible.
Client engagements are anonymized: no client names, account identifiers, or resource names are shown. Dollar outcomes live in the detailed writeups and the resume, not the headlines.