I build AI systems that run email, publishing and growth work for businesses.
A live router picks the cheapest capable model for every AI helper task. It runs daily: in one week it handled 701 tasks and moved more than 3.5 million tokens of work from Opus to the cheaper Haiku model.
See the routing benchmarkReworked customers' landing pages and offers to raise opt-in rates and sales per click. Customers kept buying from me for years.
About the email businessA public directory of AI-agent tools, each tested in a no-network lab before publication.
Read the toolwitness.ai case study| 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 | Three models | Self-hosted | MCP coordination |
Four frontier model families, coordinated over MCP.
Local inference for three self-hosted models.
Python, PHP, TypeScript, JavaScript, SQL and Bash, with Laravel and Next.js for web development.
Self-hosted Git and mail relay, backups with restore drills.
Prompt and context engineering, LLM evaluation, guardrails and mutation testing.
A daily pipeline pulls about 6,100 job postings from public hiring APIs, screens them with rules, and has LLMs read the survivors. The LLM's key claims must quote the posting word for word or they are rejected automatically.
A second pipeline writes persona-specific email copy. Its false-premise checker was tested on 5,487 real emails.
toolwitness.ai is a public directory of AI-agent tools. A scheduled pipeline starts each new MCP server in a no-network Docker lab and publishes only records that pass deterministic checks plus independent Codex and Grok review.
It also runs 56 automated newsletter sites, each emailed Monday, Wednesday and Friday. Read the case study.
A live router picks the cheapest capable Claude model, Haiku, Sonnet or Opus, for every AI helper task, from a one-line title. It runs daily: in one week it handled 701 tasks across 53 sessions and moved more than 3.5 million tokens of work from Opus to the cheaper Haiku model.
It was benchmarked blind on 100 real tasks, graded by two AI judges.
Every automated check is tested by planting a known bug and confirming the check catches it.
Mutation testing runs in development and overnight.
Code is reviewed by a different model family than the one that wrote it.
B.S. Computer Science, Business Administration minor, Southern Oregon University, 2010. Founder and Principal Engineer at IM-Empowered Inc. since 2010.
For 15 years I ran my own email and traffic business, selling traffic and solo ads to customers and ad networks.
Reworked customers' landing pages and offers to raise opt-in rates and sales per click. Owned paid media, list building, deliverability, offers and back-end resale.
AI Automation Engineer, Growth and Email. Remote preferred, Phoenix hybrid welcome.