Generation got cheap. Trust didn't.
I build the systems in between, from air-gapped inference on 100+ GPUs to the gates in AgentOps.
// what I operate across
SUBSTRATE → PLATFORM → FRONTIERI understand the stack because I came up through it.
One through-line across three layers: you can see the system, reverse a bad change, and promote without a hero.
- OPERATE THE FRONTIER
AI Infrastructure & Multi-Agent Systems
07Agents
2025–present · building in the openAgentOpsAgent output is cheap, confident, and wrong somewhere. Review by feel is how that wrongness ships.400+ stars · 50+ skills Gastown OperatorMulti-agent work needs a control plane: workspaces, coordination, lifecycle. Kubernetes already knows how to run that shape.3 ecosystem projects: Gastown · Beads · KAgent06AI
2023–2026 · AI platform at scaleAir-gapped LLM inference platformTeams needed dependable model inference in an air-gapped environment without a cloud fallback.100+ GPUs · air-gapped inference Air-gapped AI platform (reference)Standing up an AI stack in a disconnected environment means re-deriving the whole map every time: inference, vector stores, gateways, observability, agents.139 Helm templates - PAVE THE PLATFORM
Platform Engineering & Distributed Systems
05Distributed platforms
2023–2026 · AI platform at scaleGitOps at 100+ clustersA hundred-plus cluster fleet fails by drift: hand-edited configs, snowflake clusters, no way to prove what shipped where.100+ clusters · 58 applications Release automationProvisioning cycles ran weeks, and deployment rework ate the platform team: the same fixes, reapplied by hand, cluster after cluster.~80% less rework · provisioning weeks → hours - BUILD THE SUBSTRATE
Infrastructure Engineering
04Infrastructure
Stood up classified networks and led a 0-to-1 datacenter build. Wrote my first production Python automation.
2020–202303Networks
Built and managed 40+ healthcare networks across Virginia under full HIPAA compliance. The first arena where being wrong had consequences.
2016–202002Systems
Self-taught Linux, networks, and hardware. The credentials came after the practice, earned while working.
roots01Cables
Started at the help desk, turning up cables and troubleshooting the systems behind them.
roots
Every chapter, the same move one layer up. The arc in full →
// how the work holds
SLOP → TRUSTRaw model output is cheap and wrong somewhere. The gates make a subset safe to ship.
THE WORK
Brief the agent · lay the rails · gate the work · govern the loop. The agent that wrote the code never grades its own work.
// field notes
the archive →
Fire for the Rest of UsSame gatekeeper discipline that keeps agents trusted on aircraft: give context, set boundaries, check the work, keep what matches. It works on an inbox too.- The Validation Bottleneck: Why AI Output Quality Is the New CI/CDGeneration is cheap. The bottleneck is proving whether AI output is correct, safe, and worth shipping.
- When It's LyingYour AI will tell you something false with complete confidence. It doesn't know the difference. Here is how to catch it, without understanding a thing about how it works.
- Reliable AI Agents Need a DevOps Feedback LoopBound the work, check it independently, record what passed, and preserve reviewed lessons. Reliable agent delivery looks more like DevOps than magic.
// independent practice
Fire for the rest of us.
One secondary practice, with a path for teams and a path for individuals.
Got a system that has to hold?
If the problem is infrastructure, platforms, or AI that has to ship under hard constraints, write me.