The TransUnion and Lobbi work is private, so there's no code to link. I'm happy to walk through any of it.
80+ applications, $1.8M in license costs avoided
At TransUnion I led the migration of 80+ applications from Docker EE to Kubernetes ahead of a hard deadline, which avoided $1.8M in enterprise license costs. About 3 of us did it over about 18 months. I converted the Compose deployments to Helm charts and wrote a Python health-check library that backs the pod probes for about 80 deployments.
Hard deadline. We made it.
408 pipelines, 1,000+ repositories, 23 teams
At TransUnion I consolidated three separately maintained Jenkins instances onto one Harness platform, covering 408 pipelines, 1,000+ repositories and 23 development teams. OpenTofu codifies each team's access on a least-privilege model. Build times stayed about the same, and the gain was one CI system to maintain where there had been three.
Not faster. Just one thing to fix when it breaks.
Release build job: 11 to 14 minutes, now about 3.4
At TransUnion a team gets a new Harness pipeline by adding an entry to a Terraform file, without waiting on the platform team. At The Lobbi Corp I built a GitHub Actions factory of 19 reusable composite actions (shared build steps), each pinned to an exact version across 17 repositories. It cut the release build job from 11 to 14 minutes down to about 3.4.
In production beta with external testers
Wagon Works is The Lobbi Corp's AI assistant for small teams, a desktop app that works inside their Microsoft 365. It finds the file, names the source and drafts the next step, and a person approves before anything sends. I co-led the three-person team that took it to a production beta, from v1.0.0 through v1.0.28, with Azure-hosted licensing, Stripe seat billing and Entra sign-in.
The team merged 661 pull requests in about 3.5 weeks
Coding agents write the code I deliver. I design, rule on decisions, verify and ship. At Lobbi I designed and ran the delivery model, with parallel worktrees, coordinator and worker agents, and cheaper models implementing while stronger models review. Under it the team merged 661 pull requests in about 3.5 weeks, and I merged 103 of them.
The agents write. I answer for it.
Nothing publishes until its checks pass
I designed and delivered two pre-release LLM features for Wagon Works through coding agents. Task ranking accepts only schema-valid JSON, and meeting recaps publish only after every citation is checked against the transcript. I also led a security review before release and set the rule that the assistant's write tools need the person's approval once a turn has read outside content.
A release now fails on any unsigned file
Wagon Works crash-looped on fresh Windows 11 machines because Smart App Control, a Windows 11 security feature, blocked unsigned DLLs. I diagnosed it from PostHog telemetry and made full-payload code signing a release gate, so a release now fails on any unsigned first-party file.
One server, about 2 years, about 20 Plex users
I have run one Unraid server for about 2 years. Its Docker services include Plex for about 20 users and Nextcloud on MariaDB, published through Nginx Proxy Manager with Let's Encrypt certificates and Cloudflare dynamic DNS. An Ansible playbook, unraidAnsible, sets the server up, and harnessUnraid is a template for running Harness CI/CD delegates (the workers that run its jobs) on Unraid.
Plex uptime is a family matter.
Every animation rebuilds from a script
Built with Vite, TypeScript and Effect. The intro is a Blender scene generated entirely from a Python script and rendered headless, so the animation can be rebuilt from code.
Yes, I animated a toy version of myself.