Alex Villarroel
01The thesis

Industrial systems // Business operations // Guarded AI

I build systems for work that have to hold up in the real world.

I came up through industrial electrical and controls. In that world a system isn't finished the first time it runs. It's finished when you can trust it under real conditions. I work the same way on projects, businesses, software, and AI.

See the work Read the Field Notes

Industrial contractor. Project operator. AI systems builder.

  • Industrial work since 2017
  • Founder, Streamline Automation Solutions
  • Founder, Constellation Labs
  • Builder, Shop Marshal
02What I do

The three kinds of work I do.

I run real projects, I build the systems that carry the repeat work, and I write down what I learn when it holds up or when it doesn't.

The physical work

Industrial electrical, controls, networks, project delivery, estimating, planning, execution, testing, and commissioning.

The systems behind it

Business workflows, software, agent fleets, memory, review, authority boundaries, and the plumbing that lets work keep moving without the owner routing every step.

What I publish

Field notes, open systems, diagrams, experiments, and plain writeups of what worked, what didn't, and why.

03The bridge

I build AI the way I wired industrial systems.

A guardrail is a request the system can ignore. An interlock physically won't let the bad thing happen.

In a plant, you don't stop a dangerous action with a warning sign. The thing that stops it sits outside the machine and doesn't ask permission. AI that carries real authority needs the same: explicit permissions, external gates, actions you can watch, a human override, and a way to stop it.

  • InterlockExternal gate
  • PermissiveExplicit permission
  • AlarmObservable action
  • Manual overrideHuman override
  • ShutdownA way to stop

See the system architecture

The constraint lives outside the thing it constrains.A recurring principle across my controls, business, and AI work

04Proof

A few things I've built, with honest status.

A handful of pieces, each laid out the same way: the problem, what I owned, the system, the constraint that mattered, the outcome, and evidence you can check. Anything I can't confirm yet is marked illustrative. I don't publish made-up numbers.

Shop Marshal

Contractor operations, estimate to paid.

In development
Problem
The second shift. Running actual contracting work throws off a pile of office work: estimates, crews, documents, change orders, billing, cash visibility, all of it rebuilt by hand after the field day is done.
Role
Author. I defined the operating model and the single record that runs from estimate to paid. Streamline owns the product; I'm its author, not its owner.
System
Contractor operations software that carries a job from estimate to paid on one connected record.
Constraint
A contractor's money and field records can't survive a lost update or a silent mis-book. So the one record is the source of truth, not a dashboard sitting on top of a dozen scattered tools.
Outcome
In private early access, in development. Never described as live or launched.
Evidence
shopmarshal.com, the early-access product page.

Constellation Labs

The architecture behind coordinated AI systems.

Research
Problem
Coordinated AI systems need an architecture you can pressure-test, not a demo that works once on stage.
Role
Founder.
System
The private research and architecture lab where memory, review, operating rooms, and authority boundaries get built and hardened.
Constraint
Before an autonomous system runs, it has to answer one question: who can stop it. That authority gets built in from the start, not bolted on later.
Outcome
An active research lab. What gets proven here is what the rest of the work is built on.

Streamline Automation Solutions

Guarded AI operations for businesses the owner still runs.

Operating
Problem
In a business the owner runs, the owner becomes the human router sitting between every tool, every person, and every payment.
Role
Founder.
System
The company that installs guarded AI operations into owner-run businesses, so the repeat work moves without the owner handing over control.
Constraint
The owner keeps the call on anything with real consequences. Automation does the legwork, not the deciding.
Outcome
An operating company delivering guarded AI operations.

Orchestrator OS

An open framework for running AI as a system.

Open source
Problem
AI work stitched together from loose prompts has no system behind it: no roles, no rules, no review, no memory that survives a handoff.
Role
Author.
System
An open framework for coordinating agents, context, tools, ceremonies, memory, and review as a system rather than a collection of prompts.
Constraint
Open and inspectable. It has to make sense to someone who didn't write it.
Outcome
Public technical proof, open source.

Industrial project work

Field and project delivery, scope through commissioning.

Completed
Problem
Manufacturing plants and municipal water and wastewater systems need the drawing, the installation, the logic, the network, the schedule, and the people to all agree before startup.
Role
Contractor and project lead across estimating, labor, materials, controls, electrical systems, documentation, and commissioning.
System
Field and project delivery carried from scope through installation and commissioning.
Constraint
Outages, budgets, schedules, safety, old equipment, and real people set the limits. It's only finished when it holds up under real conditions.
Outcome
Selected commissioned projects. Illustrative until specific figures and customer details are approved for publication.
Evidence
Customer-approved photographs, one-line drawings, and scope excerpts. Illustrative sample only; real artifacts shown where a customer has approved.

AI Prompt Sync

A practical public tool from the larger work.

Operating
Problem
Rebuilding prompt and working-directory context by hand every time you open an AI conversation.
Role
Author.
System
A practical public tool for moving prompt and working-directory context into AI conversations without rebuilding the operating state each time.
Constraint
Free and public, kept on its own, backed by a separate entity so a free tool that also takes donations stays walled off from everything else.
Outcome
A live public tool, free to use.
05Now // August 2026

What I'm working on right now.

Right now I'm building Shop Marshal toward early access, working on the architecture behind Constellation Labs, installing guarded AI operations through Streamline, and still running real industrial project work. The thread through all of it is cutting down the handoffs that depend on the owner, without taking away human authority.

See the current build log

Last updated August 15, 2026.

06Field Notes

What I write about here.

I write about industrial systems, running projects, businesses that lean too hard on the owner, guarded AI, agent architecture, and the lessons that only show up after a system meets real conditions.

The first notes are coming soon.

Get Field Notes by email

No content treadmill. I only send one when I've got something actually worth passing along.

07Where the work lives

How the companies fit together.

This site covers the common discipline. Each company or product does something narrower.

Streamline Automation Solutions

Streamline installs guarded AI into a business, so the repeat back-office work gets done across the tools the company already uses. The owner still keeps the call on anything with real consequences.

Work with Streamline

Constellation Labs

Constellation Labs is Streamline's private research and architecture lab. It's where the architecture behind the products gets built and proven before it ships.

Enter Constellation Labs

Shop Marshal

Early access / in development

Shop Marshal is a Streamline product, built in the Constellation lab. It's contractor operations software that carries a job from estimate to paid on one connected record. It's in private early access, still in development.

Explore Shop Marshal

Public systems and tools

Orchestrator OS is an open framework for coordinating AI agents, context, tools, ceremonies, memory, and review as one system, instead of a loose pile of prompts.

AI Prompt Sync is a practical public tool that moves your prompt and working-directory context into an AI conversation, so you don't rebuild the operating state every time.

See the open framework Try AI Prompt Sync

08Background

The field is the foundation, not the ceiling.

I started in industrial electrical and controls in 2017, inside manufacturing plants, municipal water and wastewater systems, and other places where the drawing, the install, the logic, the network, the schedule, and the people all have to agree. From there I moved into estimating, project delivery, commercial responsibility, company operations, and software, without losing the field. That's why I think in interfaces, failure modes, handoffs, authority, evidence, and commissioning.

Read the full background See selected industrial work

Alex Villarroel
Alex Villarroel. Texas.
09The work is not the whole life

There's more to it than the work.

I'm a husband, a father, and I train Brazilian jiu-jitsu. Family, health, training, time at home: none of that is separate from how I think about the work. It's the reason I care about building things that don't need one person holding every system together forever.

  • Controls and automation work
  • Brazilian jiu-jitsu training
10Choose the right path

Tell me what you're working on.

Need commercial implementation? That's Streamline. Contractor operations, that's Shop Marshal. Systems architecture and research, that's Constellation Labs. For partnerships, advisory work, speaking, media, or just a direct conversation, reach me here.

Work with Streamline Explore Shop Marshal Contact Alex

Tell me what you're working on and I'll point it the right way. Your name, your company if you have one, why you're reaching out, and a link or two if it helps.

Email me

Or write me directly at me@alexvillarroel.com.