Solutions Architect · Platform Engineer · DevSecOps
I get shit done.
Twenty-five years of it. Cloud architecture, developer platforms, CI/CD — and lately AI agents pointed at operational work, held to the same bar as everything else I ship. Mostly the problems that have already defeated everyone else: the outage nobody can trace, the migration that stalled twice, the system three people quit over. That's the call I take.
- Years shipping
- 25+
- Primary domain
- Cloud & platform
- Called in for
- The hard one
01 — Focus
What I actually do
The work spans architecture through operations. These five areas are where I spend most of my time.
-
AI & agent systems
Leading development of an AI-driven data platform, and building agents aimed at operational work — triaging issues, investigating cluster problems, answering questions against real documents. The interesting parts are the unglamorous ones: routing cheap models at cheap tasks and expensive ones only where they earn it, retrieval that grounds an answer in a citable source, classifying every command an agent wants to run so the destructive ones need a human, and evals so a prompt change can be measured instead of guessed at. Non-determinism is a reliability problem, and in regulated environments it gets treated like one.
-
Cloud architecture
Designing systems that survive contact with production — account and network topology, multi-region and failover strategy, and honest cost modeling before the bill teaches the lesson instead.
-
Platform engineering
Golden paths, service templates, and self-service infrastructure. Treating other engineers as the customer, and measuring the platform by how much toil it removes rather than how much of it exists.
-
CI/CD & automation
Pipelines that are fast, legible, and boring. Reproducible builds, progressive delivery, GitOps, and infrastructure as code that a stranger can read on a Tuesday afternoon and understand.
-
DevSecOps
Security built into the pipeline rather than bolted on at review time — secrets management, supply-chain integrity, hardened baselines, and gates that block real risk without blocking real work. Much of it in strict federal environments — FedRAMP compliance, workloads in GovCloud, STIG-compliant baselines. That work runs on different rules than commercial: the control set is prescribed rather than chosen, and evidence has to be continuous instead of assembled the week before an audit.
02 — Work
Selected projects
Client work up front, then things built to scratch an itch. Source for the open ones lives on GitHub.
-
Government platform
GovOrch AI
An AI-driven data orchestration platform for government — data catalog, visual pipeline builder, and natural-language questions answered from the data itself. I lead application development on my team and drive DevSecOps: the delivery pipeline, the security gates, and the controls a federal deployment has to satisfy.
Private — no public repository -
TypeScript
shipwright
A Kubernetes SRE agent that reasons about cluster problems with Claude on Bedrock. Every command the model wants to run is classified first — read-only runs, disruptive needs a human, destructive is refused — with unrecognized commands denied rather than allowed. The classifier enforces it, not the prompt, so a persuasive model can't talk its way past it.
View repository -
TypeScript
code-monkey
A GitHub issue-triage agent on Bedrock, built around a separation I keep coming back to: the model that writes is not the model that reviews, and a controller sits above both. Haiku classifies, Sonnet drafts and critiques. The orchestration, infrastructure, and schema are built; the autonomous execution path is still being wired.
View repository -
Groovy
ci-fast
Jenkins pipeline tooling in Groovy, aimed at the single most underrated developer-experience metric: how long you wait for a build.
View repository -
Python
cat-feeder-5000
An automated pet feeder in Python. Physical hardware, real consequences, and a very unforgiving set of end users.
View repository -
Python
codex
RAG document Q&A — upload PDFs, ask questions, get answers streamed back with citations to the source passage. Chroma for vectors, a chunking and retrieval pipeline, and an eval harness that scores answer quality so changes to the prompt can be measured instead of guessed at.
View repository -
HTML
cssr
An internal web app for recording student course scores — small, focused, and built for the people who had to use it every day.
View repository
03 — Stack
Tools of the trade
Opinions held loosely, but held. This is what I reach for by default.
Cloud & infrastructure
Languages
Delivery
Operations
AI & agents
Security & compliance
04 — Contact
Let's talk shop.
Architecture reviews, platform work, or a thorny infrastructure problem that has outlived its last three fixes — happy to hear about it.