I run the IT behind a live international trade show, build the AI tooling my team uses every day, and manage this site with a workflow that records and reverses every change. These three projects show how I work: understand the problem, build carefully, and leave a trail anyone can follow.
01 · Live-event infrastructure
American Film Market IT operation
The American Film Market is one of the largest film business events in the world, a marketplace where thousands of producers, distributors, sales agents and financiers from more than 70 countries finance, license and sell independent films. It has run since 1981, and I have led its IT operation for 23 consecutive years. The build takes months, the deadline never moves, and there is no second take.
What I did
- Own the annual IT plan: inventory, timeline, staffing, vendors and the runbook.
- Size and stage equipment for about 20 year-round staff and 50 to 60 peak users, with a seasonal ramp from July.
- Lead a whole-hotel conversion: guest rooms become staff and exhibitor offices, plus a rented theater for the screenings.
- Define network requirements with hotel and theater IT in facilities we do not own, working around constraints like ballrooms and third-floor staff rooms sharing one subnet.
- Author the annual golden laptop image that the rental vendor clones, so each user can deploy in minutes.
- Lead the office-to-hotel cutover and run the systems for the duration of the show, directing a team of three and coordinating the Admin and Production departments without formal authority.
Tools and technologies
- Network design, firewalls, VLANs and routing for a temporary show floor.
- Mass imaging and deployment from a single golden image.
- Vendor and venue coordination, runbooks and documentation discipline.
- A recent run of venue changes: the Loews Santa Monica Beach Hotel to 2022, Le Méridien Delfina in 2023, the Palms Casino Resort in 2024 and the Fairmont Century Plaza since 2025, each needing a new design on the same deadline.
Why it matters: The cloned image lets each user deploy in minutes, and the runbook carries more of the load as a smaller crew delivers the same build with less on-site preparation. When the venue changes, the plan is what changes with it.
02 · Team automation
AI tooling for the MSP team
I work at a managed service provider that supports roughly 400 to 500 seats across real estate, legal, nonprofit and film clients. Investigating a single alert can span ticketing, remote monitoring, endpoint security, Microsoft 365 and backups. I wanted my team to use Claude across that whole stack, safely.
What I did
- Led the rollout of Claude to the tech team and built every MCP server, agent skill and integration behind it.
- Built five production MCP servers: Microsoft 365 Management (tenant admin with preview-then-execute), M365 Security Investigation (read-only sign-in, audit and mailbox forensics), N-sight RMM (read-only device, check, patch and backup data), WatchGuard EPDR (endpoint protection, security events, risk and patch posture) and Freshdesk (ticket triage, private notes and knowledge base).
- Built the agent skills the team uses daily: Freshdesk ticket triage, an M365 breach report, safe Google Workspace editing, and a morning brief and inbox sweep.
- Built automation pipelines: an n8n phishing triage flow, a weekly backup review that gathers the week’s backup alerts and flags the jobs to inspect, an AppSheet approval and filing app, and a Composio integration that gives each tech’s Claude account API access across the stack.
- Built a vendor assessment engine in R with local embeddings, so client data never leaves the machine. It drafts answers to supplier-control questionnaires from a knowledge base, cites the evidence and flags anything it cannot support.
Tools and technologies
- Model Context Protocol (MCP) servers, most read-only by design.
- Claude agent skills for investigation, documentation and recurring reviews.
- n8n, AppSheet and Composio for pipelines and integrations.
- R with local sentence embeddings for the vendor questionnaire engine.
Why it matters: The team can use AI for investigation, documentation and recurring reviews inside the tools it already runs. Ticket triage is the most-used skill, the weekly backup report gives us specific jobs to examine, and the vendor engine never claims a control it cannot evidence.
03 · Reversible publishing
This AI-managed WordPress site
This site is both my personal site and a working lab. Several AI agents build and change it, so I needed one shared process that records who did what and always leaves a way back. WordPress runs in Docker on a Linux host, reachable only over my private Tailscale network.
What I did
- Manage the site with Claude, Codex and OpenCode over SSH and WP-CLI, plus a WordPress MCP adapter for agents that only have API access.
- Take a database and content snapshot before every change, verify the result, then export the site and commit it to a private GitHub repository.
- Publish a numbered Lab Note for every working session, tagged with the harness and closed with a change record and the exact rollback command.
- Track each agent’s process, snapshot discipline, typed commits and rollback rate on a Harness Scoreboard that rebuilds from Git.
- Ran an accessibility audit against WCAG 2.0, 2.1 and 2.2 AA with axe-core and fixed what it found, including a visible keyboard focus outline, a proper H1 on the Lab Notes page and landmark fixes.
Tools and technologies
- WordPress in Docker on a Linux host, with WP-CLI and the WordPress MCP Adapter.
- Tailscale for private access and GitHub for tracked, readable content exports.
- Snapshot and rollback scripts, and a Python scoreboard generator.
- axe-core for the accessibility audit and Git history for the process metrics.
Why it matters: A change, commit, rollback and verify round trip has been tested, so any experiment here can be undone. The Lab Notes keep the process visible, including what failed, and the same workflow now runs a bake-off that compares AI agents on one task.