Deep systems that automate work, orchestrate AI, and integrate your stack.
Multi-agent AI systems
Crews of LLM agents with human approval gates — I run five in production, around the clock, on my own cluster.
Business automation
APIs and services composed into event-driven workflows that remove manual steps — the kind that took an intake process from eight days to under one.
Self-hosted platforms
Production infrastructure on your own hardware — privacy, control, and resilience as design constraints. I run 50+ services this way.
Discord & API integration
Bots, webhooks, and API bridges that keep people and systems in the same conversation — my agents post their standups to Discord twice a day.
02 · How systems grow
Every platform starts as one page. Architecture is what happens next.
Scroll to watch a single-page application become a multi-service platform —
the same progression I design for deliberately: separation of concerns, a
dedicated data tier, focused services, an orchestrated network, replication
that survives node loss, and a delivery pipeline that ships it all. Each
stage is annotated with a company that famously lived it.
00 · SINGLE-PAGE APP
One page, one deploy — everything lives in the browser.
Instagram launched this way in 2010 — a Django monolith that reached 14M users with three engineers.
Architecture is only half the story. The other half is
who builds it — and how that group grows.
03 · How teams grow
Systems don't scale alone. The organization around them grows too.
Step from a solo developer shipping straight to production, through
review gates, squads with real ownership, release engineering, and change
management — to governance enforced by the pipeline itself. It's the
human architecture I've built and led inside, from startup pace to
enterprise scale.
00 · SOLO DEV
One person, full context. Commit, deploy, done — the pipeline is muscle memory.
Linux began as one developer: Linus Torvalds posted v0.01 himself in 1991.
04 · The toolkit
Built on Docker, Linux, and open standards — designed to integrate deeply.
Hover to inspect · click a module to expand its parts.
05 · The physical substrate
A proper 3-node HA cluster — asymmetric by design, not accident.
A 3-node Proxmox cluster with deliberately asymmetric roles: a Dell R720
workhorse carries all the compute, an edge box holds only the network
control plane — so the network survives the compute host going dark —
and a Raspberry Pi casts the tie-breaker quorum vote while running
Pi-hole DNS for the LAN. One $50 Pi; two real jobs.
Hover to inspect · click a node to expand its parts. Hover a hypervisor header or the Pi to see the quorum ring.
06 · How systems flow
Agents execute. Humans approve. Systems ship.
INTAKE — Work enters as cards, briefs, and events.
07 · Selected projects
Three systems built to solve concrete problems — all running in production today.
Hermes Multi-Agent Crew
Five always-on LLM agents that pull work from a kanban board, run standups, and open pull requests — a human holds the merge button.
Agents
5 always-on · own model + cron each
Pipeline
Vikunja kanban · Discord standups
Models
Routed per agent — one switch cut cost 10×
Gates
PR-only — a human merges everything public
5-Node Self-Hosted Platform
Production infrastructure running 50+ services on 5 Swarm nodes across 2 Proxmox hypervisors — media, photos, automation, monitoring, and this site.
Compute
5 VMs · ~36 vCPUs · ~41 GB RAM
Storage
Dedicated NFS server · 3 shared exports
Sync
Configs replicated across nodes every 10 min
Monitoring
Prometheus + Grafana · Uptime Kuma
PR-Gated Publishing Pipeline
This site. Git-backed content with CI quality gates and human approval — AI agents draft, a human merges, merge is publish.