How I build with AI
The resume page is the short version. This page has the interactive parts: the computers I work on, how work gets routed to the right model, and the private AI stack I design for customers.
An agent platform I run every day
Four computers running Claude Code with version controlled memory, written operating rules the agents follow, local first model routing, and hooks that enforce code quality. It is how a customer conversation becomes a demo, a sizing, and a deck in about a day and a half.
Four computers, one identity. Memory is git plus a search index, so every architecture decision and rejected approach is searchable months later.
Every prompt is classified and sent down a cost ladder: local GPU first, cloud only when it earns it. Type something and watch where it lands.
Waiting for input.
The private AI stack I design for customers
The layered model behind the 2026 engagements. Private by default: no cloud, no outside API keys, data never leaves the customer's network.