enterprise ai architecture & advisory · rochester, mn
16+ years architecting distributed cloud systems and mission-critical AI — moving generative AI from fragile prototypes to resilient, HIPAA- and HITRUST-compliant platforms on GCP, Rust, and Go.
three things
Ship agents that hold up in production.
tool loops · evals · humans in the loop
Build on infrastructure auditors accept.
HIPAA · HITRUST · in-tenant by default
Leave a team that can run it without me.
enablement · not dependency
how the work thinks
An agent that works is an agent whose control loop you can draw on a whiteboard. The interesting parts are the state machine around the model calls, the tools you chose to expose, and the eval harness that tells you whether the last change helped.
That node riding the circuit? It is bound to your scrollbar — scroll up and it runs the loop backward. The same engine drives everything moving on this page.
A prompt chain drifts by the fifth turn; a state machine can't. Every transition is enumerated, every tool call is gated, and an auditor can read the one path the system is allowed to take — which is the actual argument for shipping agents into a regulated environment.
what I help with
all services →selected work
the full index →cli · rust

code review (rust)
Multi-agent code review on rig.rs with a standalone four-tier diff compressor.
75% token reduction on 2k-line diffs, backed by 143 tests.
2026agentsread case study →
service · rust
orca
Agentic platform for healthcare workflows — Rust core, Vertex AI, human-in-the-loop by default.
Production-shaped: human-in-the-loop by default, guardrail pipeline on every tool call.
2026agentsread case study →
case study
verve
Full-stack Zig web framework (v0.5.0). Pure-Zig, zero deps, SSR with fine-grained reactivity, per-island WASM code splitting, a GSAP-class animation engine, an in-house WebGL2 3D renderer, and single-binary deploy.
Zero third-party dependencies, 731 tests, ~73 B per-island WASM overhead.
2026libraryread case study →
recent writing
all posts →- 2026-06-137 minessay
give the agent the harder job
The cost of asking an agent to attempt something has collapsed. Most people still scope their asks to what they already believe is possible — that is the mistake.
read →
- 2026-05-217 minessay
why I write Zig
Zig is the language I reach for when I want a Go-shaped operational profile, Rust-shaped compile-time honesty, and a stdlib that does not lie to me about IO.
read →
- 2026-04-227 minessay
a finops mindset for small teams
Every architectural decision is a bill waiting to arrive. The small team that knows which is the one that keeps shipping.
read →