Featured Projects by Claude with Opus 5.5 High

The problem

The American Film Market is one of the largest film business events in the world. It isn’t a festival, it’s a marketplace: thousands of producers, distributors, sales agents and financiers from more than 70 countries come to finance, license and sell independent films. It’s run by IFTA, the Independent Film & Television Alliance.

All of it runs in facilities we don’t own. Guest rooms become staff and exhibitor offices, and screenings run in a rented theater. The show moved three times in three years, from the Loews Santa Monica to Le Méridien Delfina, then the Palms in Las Vegas, and now the Fairmont Century Plaza, while on-site prep time, overtime and crew size were all cut.

What I did

• Own the annual IT plan: inventory, timeline, staffing, vendors and the full runbook
• Size and stage equipment for about 20 year-round staff and 50 to 60 peak users, with a seasonal ramp that starts in July
• Define network requirements with hotel and theater IT, and engineer around constraints like ballrooms and third-floor staff rooms sharing one subnet
• Author the golden laptop image the rental vendor clones, so each user deploys in minutes
• Lead the office-to-hotel cutover, run systems for the show and direct a team of three

Tools and technology

• Network design for venues run by someone else’s IT team
• A golden laptop image for mass imaging and deployment
• A documented runbook covering inventory, timeline, staffing and vendors
• Coordination with the hotel, the theater, the rental vendor and the Admin and Production departments

Why it matters

The deadline never moves, so the planning carries the project. Each venue move meant a new network design, a new office layout and a new venue IT team, all on the same fixed date. With a smaller crew and less time on site, the runbook is what makes every hour count.

It’s the clearest example of how I work: fixed-deadline planning, vendor and venue coordination, and leading people I don’t formally manage.

The problem

I’m the technical lead for a four-person team supporting roughly 400 to 500 seats across real estate, legal, nonprofit and film clients. The work lives in many tools: Freshdesk for tickets, N-able N-sight for RMM, WatchGuard EPDR for endpoint security, Microsoft 365 and Entra ID, and our backup platforms.

An AI assistant is only useful here if it has structured, scoped access to those tools, and only safe if it can’t change things it shouldn’t.

What I did

• Led the rollout of Claude to the tech team, including every MCP server, skill and integration behind it
• Built five production MCP servers: Microsoft 365 Management, M365 Security Investigation, N-sight RMM, WatchGuard EPDR and Freshdesk
• Wrote agent skills, including Freshdesk ticket triage (the team’s most-used) and an M365 breach report that turns an affected account and a rough timeframe into a structured investigation
• Built automation pipelines: n8n phishing triage triggered by a Freshdesk webhook, and a weekly backup review of backup alerts
• Run seven scheduled agent jobs on a self-hosted lab with Hermes and OpenClaw agents

Tools and technology

• Claude and the Model Context Protocol (MCP)
• n8n, Composio and AppSheet
• Freshdesk, N-able N-sight, WatchGuard EPDR, Microsoft 365 and Entra ID
• Self-hosted agents and local models on a Linux host reachable over Tailscale

Why it matters

The triage skill investigates alerts across Microsoft 365, endpoints, EDR and RMM, writes a clean private note and closes informational tickets. The weekly backup review collects the week’s backup alert tickets, generates a report and flags the jobs that need a closer look.

Every agent follows one design rule: read-only by default, and nothing changes without a way to undo it. Most of the MCP servers are read-only, and the one that administers Microsoft 365 tenants previews each change before it executes.

The problem

I wanted a personal site that doubles as a lab: a place to see how different AI harnesses handle real changes. That only works if every harness follows the same process and leaves the same paper trail, and if any change can be reverted no matter which tool made it.

What I did

• Self-hosted WordPress, MariaDB and WP-CLI in Docker on my own Linux box, reachable only on my private Tailscale network
• Gave agents two control planes: SSH with WP-CLI for full control, and the WordPress MCP Adapter with a small custom abilities plugin
• Required a snapshot before every change, and put the site under Git: each sync exports pages, posts, templates, menus, settings and a restorable database dump, then commits and pushes
• Made rollback work to any commit, with every rollback recorded as its own commit, plus a 15-minute job that captures changes made through the MCP or wp-admin
• Wrote one shared set of rules every harness reads, and a numbered Lab Note for every working session
• Built a Harness Scoreboard computed from the Git history, and a bake-off that gives each harness the same task and rules

Tools and technology

• WordPress with the Ollie block theme, MariaDB and Docker
• WP-CLI, the WordPress MCP Adapter, Git and a private GitHub repo
• Tailscale for private access, plus shell and Python scripts for snapshots, exports, rollback and scoring
• AI harnesses including Claude, ChatGPT, Codex and OpenCode

Why it matters

It puts the same design rule into practice: nothing changes without a way to undo it. Every session that changes the site ends with a Lab Note that says what worked and what didn’t, and the scoreboard shows how well each harness follows the process.

When an accessibility audit found a keyboard focus problem, the fixes went through the same snapshot, change, verify and commit loop, and a re-scan showed zero violations on every page.