A practice, not an agency.

I trained and work as an aerospace engineer. Aircraft 6-DOF simulation, flight dynamics, control law development, avionics, and the verification work that has to survive review by people whose job is to find what you missed. In that field a model that produces plausible numbers is not an achievement — it is a liability until you have shown why the numbers are right.

That is the habit I brought to AI systems, and it turns out to be the thing most of this industry is missing. The hard part of an agent is not getting it to respond impressively. It is defining what correct means for the task, instrumenting it so you can tell when it drifts, bounding what it is allowed to touch, and deciding which actions a human still has to sign off on.

I also build and operate my own agentic infrastructure — a fleet of specialist agents on a declaratively managed Linux host, with persistent memory across sessions, scheduled autonomous work, voice control integrated into the desktop itself, and approval gates on anything that leaves the machine. The film on the homepage is that system, recorded live. It is not a product I sell. It is where I find the failure modes before they show up in a client's project.

Work comes in three shapes: building AI applications, integrating agent harnesses into existing engineering environments, and implementing workflows that people currently run by hand. The common thread is that something has to actually be in production at the end.

How I work with clients

You deal with me directly, start to finish. Engagements begin small — usually a fixed-fee audit — so you can judge the work before committing to a build. Everything I produce goes into your repositories under your ownership: code, prompts, evaluations, documentation. No proprietary runtime you have to keep paying me to license.

Client work stays confidential by default. It does not become a case study, a screenshot, or a conference talk. The only system I demonstrate publicly is my own.