I'm Dustin Winkler. I build AI infrastructure that explains itself.

25 years of production engineering. The last 2 proving that AI systems can be governed, measured, and handed off to the next person in the chair — with the receipts to prove it.

View resume → See the work ↓
The Evidence

Not claims. Results.

SSH Lockout Recovery
90 min recovery, zero data loss, 4,000 miles away
Read case study →
Doctrine Enforcement
256 assertions, 0 failures, one-line fix
Read case study →
Claude Code Contributions
3 issues filed on anthropics/claude-code
Read case study →
SQL Protection Audit
Zero backup coverage found across 4 asset types
Read case study →
API Voyager
100+ daily alerts reduced to actionable intelligence
Read case study →
Naming Convention
Solved Azure's conflicting naming constraints with math
Read case study →
Teaching

Things I've taught

Internal presentations on AI methodology delivered to 200+ engineers at a Fortune 500 enterprise.

Not a Dev, Not a Problem → Parenting Young Sheldon →
View all talks →
How it works

Build. Validate. Explain.

Build. Validate. Explain. THE CONSULTING FRAMEWORK Build Custom AI tooling for your specific environment. Hooks, agents, knowledge pipelines, automation scripts tailored to your stack. // not templates. // built for your infra. Validate Pre-deploy test suites that catch silent failures. 256 assertions, 0 failures. Measured token counts. Real output, not estimates. // if it's not measured, // it's not validated. Explain A knowledge layer so the next person can ask the platform what it does. Class A entries, decision logs, session journals, gate reasoning. // the system explains itself // to the next engineer. Each engagement leaves behind a system the next engineer can ask. // sequential, not independent. each tier depends on the one before it.

01Build

Custom AI tooling for your specific environment. Hooks, agents, knowledge pipelines, and automation scripts tailored to your stack. Not templates.

02Validate

Pre-deploy test suites that catch silent failures. Every claim is backed by measured output: token counts, byte sizes, assertion results. If it's not measured, it's not validated.

03Explain

A knowledge layer so the next person can ask the platform what it does. Class A entries, decision logs, session journals, gate reasoning. The system explains itself.

Azure Local Azure Arc Bicep / IaC Rubrik VMware Claude Code
Dustin Winkler

25 years of infrastructure. The last 2 building AI-orchestrated operations in production. Currently the only practitioner publishing governance frameworks, ebooks, and production incident journals for Claude Code on Azure.

GitHub · LinkedIn · Resume