Infrastructure I have designed and built, with architecture diagrams, and other software I have built. The first four are the best picture of how I work; three of them are open source.
The platform I set up for small software teams on bare metal: a core server running CI, GitOps deploys and monitoring with alerts; separate dev and production servers, each with a deploy agent and a secrets agent, joined by an encrypted overlay network behind a reverse proxy with automatic TLS; encrypted backups from production. Everything in Docker and defined as code.
A Go agent for servers inside networks you don't control. It only dials out, over WireGuard, and never opens an inbound port. Allowlisted commands with no shell, GitOps apply with health checks and automatic rollback, signed self-updates that can undo themselves, and Prometheus metrics.
View on GitHub
One Firecracker microVM per user session, booted in about 3 seconds and destroyed on exit or idle, to run an AI coding agent safely. Layered isolation: jailer, cgroups v2, a custom seccomp filter and an nftables egress allowlist that lets out only DNS, the AI API and package mirrors.
View on GitHub
Keeps the main branch of several repositories green. When CI turns red, a controller sends one headless agent into a clean worktree to fix it, then trusts only the CI verdict on that fix. Built after the naive version failed with several agents and people pushing to the same branch.
View on GitHubCareer development for restaurant teams. Frontline staff see the roles above them, what each one requires, and a plan that tracks material, quizzes, certifications and one-to-ones toward the next step. Mobile-first, multi-tenant, in production.
Push-to-deploy for small software teams, on the team's own machines. A control plane, an agent on every server, GitOps, automatic HTTPS, monitoring and backups.
Scheduling, talent and team communications for quick-service restaurant franchises. A constraint solver drafts shifts that comply with labor rules, delivered as a web console and an iOS app.
From news feed to published post. Ingests stories, clusters and ranks them, drafts platform-specific copy, renders branded image cards, routes everything through a human approval queue and schedules publication.
A desktop platform for natural-language software development. A fleet of named agents works on a project, each in its own terminal and worktree, surfaced in one live dashboard.