Linux is home. Messy problems in, quiet infrastructure out — built to run hot and never drop a packet.
shane@linuxbrew:~$cat ~/career.log
I work where systems, security, and shipping meet. Twenty-five years in, I've gone from Linux boxes and Fibre Channel fabric to AI platforms and CISO war rooms — and never left the command line. I harden networks with zero-trust, automate infrastructure from bare metal to the edge, and turn rough prototypes into tools people quietly depend on. Equal parts architect, operator, and closer — most at home closing the gap between “works on my machine” and “works in production, at 3am, without me.”
Enterprise AI — the boring-but-hard kind. Turning messy inputs into trustworthy, reviewable output, governed so nothing ships unseen. The interesting parts live under NDA.
Infra · IaC
netsec-iac
Python · Terraform · React
The whole house, declared. One YAML file compiles to the firewall, the Wi-Fi, and the switches — with zero-trust segmentation, dual-WAN failover, a k3s GitOps platform, and encrypted secrets. Break glass is a git revert.
A Python compiler turns one desired-state model into firewall, AP, and switch config — plus the contract the read-only front-end renders from.
Deliberate multi-tool split: Terraform for the firewall and VMs, Ansible for the APs (no provider exists), Argo CD for cluster apps.
Zero-trust network segmentation, an internal CA, and encrypted secrets — the security posture is declared in the same model.
HA firewall pair with dual-WAN failover — the control plane is a git repo, not a vendor GUI.
AI · macOS
mimir
Swift · on-device models
Meetings that never leave the machine. On-device transcription + summarization on Apple-Silicon models, written straight into Obsidian — no bot joins the call. Extended from Minute to sync M365 calendar and bolt into an AI-assist pipeline.
On-device only: Parakeet transcription and a local Llama/Gemma model summarize on the Neural Engine — nothing is sent out.
The model emits JSON; the Markdown is rendered deterministically into Obsidian with locked file paths and YAML frontmatter.
EventKit auto-detects the overlapping calendar event at record time — organizer, attendees with RSVP, location, join URL.
Recon on tap. A headless OSINT API — DNS, whois, nmap, cert-transparency behind one authenticated endpoint — fully Terraform-provisioned and locked down to least privilege.
Wraps common OSINT building blocks — DNS, whois, nmap, cert-transparency — behind one authenticated endpoint.
Provisioned end-to-end as code: Terraform + GitHub Actions, plan on PR and apply on merge — no console clicks.
Hardened and least-privilege by default; feeds a private intelligence pipeline, not the open web.
UX · IoT
shield-concierge
Next.js · TypeScript · Kotlin
The living-room command center. A Next.js kiosk for an Android TV plus a Kotlin companion — with hand-built D-pad spatial navigation (the hard part), Home Assistant control, and per-room screens.
Three locked room layouts — resort cabana, game-room marquee, living-room concierge — each with its own tiles (weather, calendar, guest Wi-Fi QR).
Hand-built D-pad spatial navigation — the genuinely hard part on a TV with no pointer.
A blended “continue watching” row across services; Plex is the one with a usable local episode-level API.
Documents a real dead end: TV launchers expose no cloud resume API, so it ships a server + per-TV two-tier resume instead.
AI · FinOps
ai-gpu-estimator live
React
Buy vs. rent the GPUs. Models real AI-inference cost — tokens/sec, VRAM, quantization, electricity and PUE — across on-prem builds and AWS / GCP / Azure / OCI.
A three-step wizard: define workloads → GPU recommendations → full TCO cost comparison.
Models VRAM, tokens/sec scaling, bin-packing, and server-cost optimization by dynamic programming; benchmarked on Llama 70B FP8.
Data-driven: 22 GPUs with specs and pricing, on-demand + reserved cloud rates across four providers, per-use-case token params.
A refresh script pulls live provider pricing; electricity, PUE, utilization, and a 4× agentic multiplier are all tunable.