Keeping production up — bare metal through cloud-native, and every platform shift in between. We build it, break it, and operate it.
Four things, done well. No frameworks, no decks, no "digital transformation." Just infrastructure you can read, run, and hand off.
The practice runs on five internal systems — botanical codenames, all New England natives. Each separates an agnostic core from its clients: the current implementation is always the first client, never the product. And where a system stands on someone else's platform, it says so by name.
A Terraform-managed AWS estate, root module down — VPC across two AZs, ECS Fargate behind an ALB, RDS, WAFv2, KMS. Plan on PR, apply on merge, OIDC-only. Plainly: it's AWS, run the way it should be run. The value is the posture.
Role-based, idempotent provisioning that turns a fresh Linux box into a configured infrastructure workstation — one command, four target profiles. It's Ansible under the hood; ours is the roles, the profiles, and the discipline.
Dashboards live as JSON and apply by Terraform on merge; one declarative Alloy collector per host. Built on Grafana Cloud — Mimir, Loki, Grafana — and we say so. What's ours is the git discipline: if it isn't in the repo, it doesn't exist.
Per-source collectors shipping full-volume logs — DNS, flows, TLS handshakes — to Axiom for 30-day search at $0/month. Fluent Bit ships, Axiom stores; betula is the persistence, the packaging, and the GitOps around both.
A self-hosted pool of headless coding agents. Drop a job on the NAS, an idle worker claims it, does the work in a clean checkout, opens a PR. It never merges. Today's workers run Claude Code; the queue doesn't care.
The full runbook is longer, drier, and in the repo. These are the six we'd bring into a room on day one.
The person who designs the system is the person who carries the pager for it. Otherwise the design is a suggestion, not a commitment.
Least-privilege isn't a checkbox — it's the default. If a compromised pipeline can reach production, the problem is the pipeline, not the compromise.
An incident handled by a sleepy engineer following the runbook is better than a hero who remembers. Write the doc. Update it when it lies.
You cannot operate what you cannot see. Logs, metrics, traces, and a single dashboard a human actually opens. No 'we'll add it later.'
Infrastructure changes are code review. OIDC, no long-lived credentials, signed artifacts. The pipeline is the contract.
A NAT gateway you forgot about is a security problem. A forgotten log bucket is a compliance problem. Run the audit monthly, not yearly.
Built by an infrastructure operations professional with 25+ years of production experience — bare-metal data centers, 24×7 ops, single-homed environments where every decision had physical consequences.
AWS / Terraform / ECS Fargate. Solidago Platform. Moving the discipline without losing the rigor.
24×7 ops. On-call rotations. Incident command. Migrated a regulated workload through three datacenter transitions without a customer-visible outage.
Bare-metal. Single-homed environments. Every change had physical consequences. Learned what 'production' actually means.
First rotation. First outage I caused. First runbook I wrote. Everything since is a refinement.
Short engagements, long ones, and one-off audits. If you know what you need, send the repo. If you don't, send the symptoms.