I build and run agent systems.
A production multi-agent system with 19,000+ commits since March 2026. Automated evaluation gates every change.
Multi-agent coordination over MCP.
| Provider | Tool or model | Hosting | System |
|---|---|---|---|
| Anthropic | Claude Code | Frontier | MCP coordination |
| OpenAI | OpenAI Codex | Frontier | MCP coordination |
| Gemini | Frontier | MCP coordination | |
| xAI | Grok (xAI) · Grok Bot | Frontier | MCP coordination |
| Open-weight | Four models | Self-hosted | MCP coordination |
What I actually run.
Orchestration
Gemini 3.1 Pro Preview and Gemini 3.8 Flash for the bulk of the work; testing Gemini 4 Argon via API.
New models go to work the week they ship. Latest: Gemini 4 Argon, in API testing since October 2026.
Local inference
Local inference on a 512 GB Mac Studio.
Languages
Python, PHP, TypeScript, JavaScript, SQL and Bash, with Laravel and Next.js for web development.
Infrastructure
Self-hosted Git and mail relay, backups with restore drills.
Model work
Prompt and context engineering, LLM evaluation, guardrails and mutation testing.
Daily data and email pipelines.
A daily pipeline pulls about 6,100 records from public APIs, screens them with rules, and has LLMs read the survivors. Answers must quote their source word for word or they are rejected automatically.
A second pipeline writes persona-specific email copy and checks every draft for false claims, proven across 5,487 emails. Daily sends to a list of about 10,000 run on my own browser automation.
A six-agent team on xAI's Grok Bot is the live-browser layer. It is fed jobs by email automatically, and its replies are read back by API. One feed queued 66 batches: 387 story fact checks and 55 site reviews.
Routing tasks by cost.
A router sends each sub-agent task to the cheapest Claude tier, Haiku, Sonnet or Opus, by five-vote consensus. It uses a one-line title so prompts and code never leave the machine.
On a 30-task scorecard, the aggressive setting cut list-price cost from $129.30 to $53.22.
How I ship.
Checks are proven before they are trusted.
Every automated check is tested by planting a known bug and confirming the check catches it.
Mutation testing
Mutation testing runs in development and overnight.
Code review
Code is reviewed by a different model family than the one that wrote it.
Founder and engineer for 16 years.
B.S. Computer Science, Business Administration minor, Southern Oregon University, 2010. Founder and Principal Engineer at IM-Empowered Inc. since 2010.
I ran my own email marketing business for 15 years, from 2010 to 2025. I bought paid media, built and monetized niche email lists, and managed acquisition, deliverability and offers.
Open to roles
Senior/staff AI engineering, agent platform and LLM evaluation roles; agent systems, growth and email automation. Remote preferred, Phoenix hybrid welcome.