ENGINEERING / SRE / AUTOMATION / PLATFORMS
More than two decades building the systems behind the experience.
I’ve built enterprise secrets platforms, operated observability at scale, and automated deployment. Today I lead a professional services team at HashiCorp (an IBM Company), and keep building: reliable infrastructure, AI systems, and tools that make complex work usable.
Utah, USA Open to interesting conversations
Explore the layers of my engineering practice.
01 / ENTERPRISE EXPERIENCE
My technical foundation was built inside enterprise systems. Reliability, security, and the people operating them have always been part of the work.
ADOBE / 2016–2022
Technical lead for Adobe’s multi-tenant Vault service. I re-architected the deployment for resiliency, planned and executed its move from data centers to AWS, and built dashboards for leadership, consumers, and administrators.
Vault · AWS · Resiliency · Worldwide training
ADOBE / 2016–2022
I architected and operated Splunk Enterprise clusters ingesting multiple terabytes of logs daily across data center and cloud, managing 6,000+ forwarders. Python automation replaced recurring manual administration.
Splunk · Python · Multi-TB/day · 6,000+ forwarders
HASHICORP / STAFF ARCHITECT / 2022–2023
At a top-5 US bank, I designed a Vault cluster and namespace vending machine. At a top-10 bank, I built identity and authentication strategy with Sentinel policies enforcing secret schemas and validation at write time.
Vault · Identity · Sentinel · Enterprise architecture
XACTWARE / 2012–2016
I built WAR and static-content deployment tooling integrated with Jenkins and NetScaler server draining for zero-downtime deployments. Application clustering, Tomcat session replication, and consolidated monitoring supported the operating model.
Java · Jenkins · NetScaler · Tomcat · On-call operations
Earlier chapters: Java development and QA at Xactware, PHP/Drupal at Astonish Design, and freelance web development. Download my IC resume (PDF) for the full career.
Curiosity, backed
by real systems.
Nomad nodesSelf-hosted cluster
TiB of storageTrueNAS + NVMe tier
Self-hosted reposOn Forgejo
OSS projects contributed toView GitHub history (opens in a new tab)
02 / CURRENT INDEPENDENT WORK
Still building. Still accountable for the result.
Explore a focus, then open the engineering notes to see the decisions behind the build.
Carry the evidence. Keep the context.
Long-running coding sessions can lose constraints and decision evidence through repeated summaries. Longhaul carries verified session state through Pi’s context compaction, preserving user corrections, code-recorded facts, and traceable decisions.
View Longhaul ↗Read the evaluation ↗
More Pi extensions: session affinity, system pin, and date tail.
A rewritten summary can gradually drop or paraphrase the facts and instructions the next step depends on.
Record commands, exit codes, files, and workspace facts with code. Ask the model for a small structured diff, then validate quotes and evidence references before accepting it. Bound the state and retrieve retired detail from the transcript.
A published synthetic-session evaluation compares chained compactions on the same model and transcripts. It measures retained rules and values, cost, and weak spots; broader model performance remains a separate question.
AI-assisted delivery guided by engineering judgment: separate observable facts from interpretation, verify the evidence, and measure the trade-offs.
ORIGINAL INDEPENDENT PROJECT / OCT 2026
Give autonomous work an operating model.
A work-request bus and service directory for coding agents. Agents declare ownership, exchange requests, and carry conversations across stateless wake-ups; a human operator can monitor, intervene, approve, or stop the work.
Agents need explicit ownership and durable conversations when work spans services and separate sessions.
A registry, inboxes, and threads coordinate work. Disposable Nomad workers use scoped identities. Concurrency, rate, and hop limits provide backstops; an operator interface and audited actions make oversight part of the architecture.
Designing the control and supervision model alongside the automation. V1 is feature-complete and end-to-end tested, with hands-on deployment testing as the next stage.
ORIGINAL INDEPENDENT PROJECT / OCT 2026
A place to build. A platform to depend on.
A seven-node Nomad cluster across mini PCs and Dell PowerEdge servers. I operate the stack from workload identity and service discovery through traffic routing, storage, and delivery.
A heterogeneous cluster still needs a consistent way to schedule workloads, deliver secrets, and expose services.
Vault provides JWT workload identity; Consul handles discovery; Traefik routes services. HAProxy and Keepalived handle balancing and failover. TrueNAS provides roughly 72 TiB of usable storage plus a fast NVMe tier.
Hands-on ownership across compute, networking, identity, storage, and the lifecycle of the services running on them.
INDEPENDENT INFRASTRUCTURE / UPDATED OCT 2026
Local models. Operational discipline.
A 27B-parameter model on a three-card AMD Radeon AI PRO R9700 system. Twin vLLM engines sit behind one LiteLLM endpoint, allowing either backend to be taken out for maintenance while the other keeps serving.
Local inference needs more than a loaded model: it needs consistent API access, a maintenance path, and useful visibility.
Qwen3.8-27B uses AWQ W4A16 quantization. Two cards run the same model as separate engines in one LiteLLM group. Metrics go to VictoriaMetrics, logs to VictoriaLogs, and Grafana connects the two with click-through log correlation and three alerts.
Applying reliability practices to emerging AI workloads, from hardware constraints to serving and observability.
INDEPENDENT PROJECT / UPDATED OCT 2026
“Can you start a Minecraft server?”
Ten Go gateway services turn the homelab into model-accessible tools. A chat interface lets family members find and deploy Minecraft modpacks by asking, with orchestration, DNS, and persistent worlds handled behind the scenes.
Non-technical users should not need to learn deployment tooling to play together.
Gateways expose systems including Nomad, Vault, Postgres, and Cloudflare. The Minecraft flow finds packs across CurseForge, Modrinth, and FTB, launches Nomad job templates, manages Cloudflare DNS, and persists worlds on TrueNAS NFS.
Connecting infrastructure automation to a concrete user need through an approachable interface.
INDEPENDENT PROJECT / UPDATED OCT 2026
Better questions. Better signals.
A self-hosted metrics and logs pipeline with a one-year retention target. Prometheus provides a short scrape buffer, VictoriaMetrics stores long-term metrics, and Grafana Alloy unifies collection.
Metrics and logs have different storage needs; a single storage choice need not fit both.
Prometheus uses remote_write to VictoriaMetrics on NFS. Loki uses MinIO-backed object storage for logs. Shelly Gen 4 smart plugs bring power telemetry into the same pipeline. A one-year retention target guides storage planning.
Selecting storage and collection patterns around actual workloads, with hardware signals included in the operating picture.
INDEPENDENT PROJECT / UPDATED OCT 2026
Reliability starts before the server.
A 5000W inverter/charger with 48V LiFePO4 rack batteries supports the lab through grid interruptions. Battery-management telemetry flows into Prometheus and Grafana, bringing power into the same operational view.
Service redundancy still depends on a working power supply.
The battery-management system provides RS485 data to the monitoring pipeline. The setup is solar-ready; panel installation is future work.
Looking beyond software to the physical dependencies of availability, then instrumenting those dependencies.
INDEPENDENT PROJECT / UPDATED SEP 2026
Most of this work lives on my own infrastructure. My public contribution history is on GitHub (opens in a new tab)
02 / PUBLISHED PERSPECTIVE
Good observability turns signals into useful answers.
Co-authored with JD Goins. A practical strategy that brings log analysis, telemetry, and API and synthetic monitoring together to help teams understand Vault health, usage, and the experience of its consumers.
QUESTIONS THE STRATEGY HELPS ANSWER
04 / HOW I THINK
I enjoy the deep technical work. The point is to make the resulting system easier to live with.
Think about deployment, visibility, maintenance, and recovery while designing the system. The person on call inherits every architectural decision.
I use AI to increase what I can deliver. I own the problem framing, architecture, review, and verification: decision records, test-first development, measured evaluations, and staged rollouts turn generated code into systems I can stand behind.
A chat request that becomes a running game server is the kind of result I care about. The infrastructure earns its place by making something useful possible.
AWAY FROM THE TERMINAL
D&D campaigns, live-fire cooking, Utah trails, electronic music, and two dogs with very little respect for uptime guarantees.
05 / NEXT CONVERSATION
Have a platform challenge, an AI infrastructure problem, or a team that cares about how systems run? I’d be glad to talk.