Ask Josh

Ask me anything about my work.

Answers come from my public notes and essays. Every citation flies you into the 3D vault.

I'm a
Notebook Fly through the vault About

Current Thesis

Getting AI to work inside a real organization is the hard part.

I build systems that take in messy input, structure it, and act on it. The point is leverage inside a real organization.

Energy and AI. Philadelphia, by way of Calgary.

What this is

Focus

AI that runs inside a real workflow, with real constraints.

Method

Small experiments, written up as they happen.

Output

Notes, builds and systems that run on a schedule.

Recent entries

All entries ->
An absorber column and a compressor train, with one stage flagged.

Builds

I Gave Claude a Process Simulator. It Found a Bug in the Simulator.

I wanted to see how far an AI agent could get with real engineering software. It built a carbon capture flowsheet in DWSIM from Python, hit a wall, read the simulator's source code, and fixed it. Fifteen flowsheets, zero clicks.

•5 min read
The actual Rupture render: a copper projectile leaving a red water balloon, with the exposed water still visibly synthetic.

Experiment Log

The Balloon Test: Building With GPT-6 Astra

I asked GPT-6 Astra to build a convincing slow-motion water-balloon rupture. It built a working app, survived six critic rounds, and the water never looked like water. Here's the result and what I'd do differently.

•4 min read
Five factory machines, an MHS driver under each, one AI agent connected to all five.

Field Notes

The Universal Adapter

MHS gives an AI a common software interface and an instruction sheet for machines that already have digital controls. Once a machine has a driver, one agent can run a workflow across machines nobody programmed to work together.

•6 min read