I build software factories and the systems behind them.

Current Focus: thirdshift

Thirdshift takes GitHub issues through implementation, standards and spec reviews, CI checks, and bounded repair loops to pull requests. Its Rust engine coordinates six coding-agent harnesses and dependency-aware parallel tickets in separate Git worktrees. Failed runs preserve their work for continuation, and merging is explicitly configured.

I test the factory's behavior as well as building it. My 183-session code-review evaluation compares a second model family with fresh sessions of the same model, with the controls, results, and limitations published together.

Explore thirdshift · Read the source · Resume

Recent Impact

• Production stack maintained 4+ years Every part of a 40-person travel company - sales, operations, payments, ACH, contractor tools - runs on a platform I keep healthy and shipping every day.
• Live platform migrations, no noticed downtime PHP 5.6 → 8, MySQL 5.7 → 8.4 LTS, CentOS 7 → Rocky Linux 9, and payments across TranSafe → Monetra → NMI - all delivered with the company still open for business.
• Revenue-protecting automation Cancellation systems, payment notifications, NACHA-formatted ACH batches to M&T Bank, and financial workflows that replaced manual intervention with reliable software.
• Measurable performance gains Brought key workflows from 5–10 seconds down to under one second, alongside ongoing refactoring that improved long-term maintainability.

Tech Stack

• Core stack: Rust, Python, TypeScript/JavaScript, PHP, SQL, Linux
• Agent infrastructure: Git worktrees, dependency-aware execution, review evaluations, CI gates, and resumable workflows
• Production systems: Docker, Traefik, Terraform, MySQL, PostgreSQL, Prometheus, and Grafana
• Integrations: payment processors, SMS providers, email services, and internal platform APIs
• Engineering habits: refactoring, documentation, bug prevention, security hardening, and performance optimization

Selected Systems

Patterns I’ve delivered repeatedly in production.

• Coding agents with explicit boundaries Role-specific Manager Sandboxes with engineering review before production, per-agent credential manifests in vaulted-agent, and muxboard's default-deny session dashboard - each one plain about what it limits and what it does not.
• Automations that run without supervision Background workflows for cancellations, notifications, and routine processes that execute reliably around the clock.
• Payment and financial infrastructure Processing integrations, encrypted financial data handling, and the supporting systems that protect revenue.
• Internal platforms and admin tooling Tools for reservation management, contractor workflows, reporting, expense systems, and operational visibility.
• Legacy modernization through steady refactoring Incremental cleanup that makes an older codebase faster, safer, and easier to ship from.

Work Style

• Start from the business problem: understand what failure costs before designing the fix
• Build systems, not patches: create the mechanism that prevents repeat work
• Document while shipping: reduce fragility for future maintainers
• Refactor as part of delivery: leave the codebase healthier than I found it, every time