Start here
A plain-English tour for anyone who does not work in AI. No jargon, no demos. Every project at the bottom links to its full technical breakdown if you want the depth.
What I actually do
Most people have used AI that answers questions in a chat box, like ChatGPT. I build the step up from that: AI that does the work. These are software helpers that take real actions on their own, such as sorting through huge piles of data, fixing computer problems, or running the day-to-day admin of a business.
The hard part, and the part I am known for, is safety. An AI that can act can also do damage. So I write the strict rules for what each helper may do by itself, what it must check with a human first, and what it can never do at all, like spending money or deleting something it cannot get back. The craft is making these helpers useful enough to trust and bounded enough to be safe.
Think of it as hiring a sharp new employee. The skill is not just teaching them the job. It is writing the exact handbook of what they may and may not do, and then proving they follow it.
A few words, defined
The terms you will see around the rest of the site, in one line each.
- Agent
- An AI that can take actions to finish a task, not just chat about it.
- Agentic system
- Several agents working together, each with a defined job, aimed at one goal.
- Coordinator and specialists
- A lead agent hands work to focused helper agents, then checks and combines what they produce.
- Guardrails
- The hard rules for what an agent may do on its own, what needs a human, and what is never allowed.
The work, in plain terms
Seven real systems. One plain sentence each, then the full breakdown if you want it.
-
Production AI for a robotics program
For a robotics program, I build and run AI agents that do real production work on their own, but only inside strict, safe limits, with a person on every important call.
See the full breakdown → -
A private assistant for a solo therapist
A desktop app that helps a one-person therapy practice with notes, scheduling, and billing, while the sensitive client information stays locked on their own computer and never gets sent anywhere. It is free and open for anyone to use.
See the full breakdown → -
A command center for running projects with AI
A lead router sends each request to the right helper, and nothing that touches the real world goes through until it passes a quality check and a permission level. It is the largest of my open-source systems.
See the full breakdown → -
Two reusable skills for AI
One looks at a messy situation and finds the single most valuable thing to do next. The other breaks a big job into pieces, runs several helpers at once, and checks the result before moving on.
See the full breakdown → -
Watching markets to answer one question
A tool that quietly watches and records how online prediction markets move over time, so one question can be answered with real data instead of a hunch. By design it can only watch and record. It can never place a bet or move money.
See the full breakdown → -
A team of helpers running my car-rental back office
A set of helpers runs the admin of my car-rental business, the scheduling, the messaging, and the paperwork, so it takes far less of my time and money to operate.
See the full breakdown → -
A team of helpers launching a product
A set of helpers does the legwork of launching a small consumer product: checking the idea, finding a maker, building the store, and getting the word out, while a person signs off on anything that costs money.
See the full breakdown →
Want the technical version?
Everything above is the simple view. The real depth, the decisions, the guardrails, and the code, lives here.