I architect AI-native cloud infrastructure and put autonomous agents in charge of the boring parts. Design, provision, operate, remediate, optimize: one closed loop, with your engineers reviewing instead of firefighting.
What I do
Three practices, one goal: a platform that stays healthy without a human babysitting a dashboard at 3 a.m.
Landing zones, hub-and-spoke networking, multi-region Kubernetes, identity, and golden paths. Built as code, reviewable in a PR, reproducible in an afternoon.
Infrastructure built to host, observe, and govern AI workloads and the agents that run on them: model gateways, MCP servers, telemetry, and cost guardrails.
Autonomous agents that watch your platform, diagnose incidents, propose or apply remediation, and learn from every loop. Humans set the policy; agents do the toil.
The lifecycle
Most shops automate provisioning and stop. The value is in wiring every phase to the next so the platform improves itself between releases.
01
Architect agent
Turns requirements into reference architecture, cost models, and threat models. Produces the plan a human signs off on.
02
Builder agent
Generates and applies infrastructure-as-code through gated pipelines. Every change is a reviewable PR with a plan attached.
03
SRE agent
Watches metrics, logs, and events. Correlates signals into a diagnosis instead of paging a human with raw alerts.
04
Responder agent
Executes runbooks with policy-scoped credentials. Review mode proposes; auto mode acts; chat mode pairs with you.
05
FinOps agent
Right-sizes, reclaims idle capacity, and feeds what it learned back into the design phase. The loop closes.
Humans stay in the loop where it matters. Policy, blast radius, and approval gates are defined up front. Agents work inside those boundaries with least-privilege identities and a full audit trail of every action they take.
Autonomous operations
This is the loop that runs inside your cluster. No human paged, no dashboard opened. Just a diagnosis, a fix, a verification, and a memory entry.
Observe
Alerts, metrics, logs, and Kubernetes events stream into the agent as structured context, not raw noise.
Diagnose
The agent correlates signals, checks dependencies, and forms a hypothesis with evidence attached.
Act
It executes a runbook step with scoped credentials, or opens a PR for review when policy says so.
Verify
Health checks confirm the fix. If it did not work, the agent rolls back and escalates.
Learn
The incident, the root cause, and the fix are written to memory so the next loop is faster.
▍ Proof of work
Reference implementations you can clone, read, and run. The same patterns I bring to client platforms.
claude-ai-kubernetes-sre-loop
A GitHub Actions–driven agentic loop that troubleshoots and remediates Kubernetes health issues. Review, auto, and chat modes; modular skills for validation, health checks, troubleshooting, memory, and lifecycle hooks.
Azure-Hub-and-Spoke-Multi-Region-Platform-Example
A DRY Terraform + Terragrunt AKS platform across two regions with hub-and-spoke networking, composed from Azure Verified Modules. Validates fully offline; OIDC plan-only CI; no secrets in the repo.
terraform-aks-traefik-module
A Terraform wrapper module for deploying Traefik on AKS with the Kubernetes Gateway API. Opinionated defaults, minimal inputs.
How we work together
Every engagement is scoped, fixed-outcome, and leaves your team with code and documentation they own outright.
A deep review of your current cloud platform with a written target architecture, cost model, risk register, and a phased roadmap toward autonomous operations.
Best for teams deciding whether and how to adopt AI-driven ops.
Discuss this engagementHands-on delivery of a landing zone, Kubernetes platform, or AI infrastructure layer. Fully as code, with CI, policy, and observability wired in from day one.
Best for greenfield platforms or major re-platforming.
Discuss this engagementDesign, deploy, and tune autonomous agents against your real platform. Start in review mode, earn trust, graduate to auto. Fractional platform architect included.
Best for teams drowning in on-call toil.
Discuss this engagementLet's talk
Thirty minutes, no deck. Bring your architecture diagram and your worst on-call story. You'll leave with a concrete next step either way.
Based in Montana · Remote-first · US & international clients