Founder memo
Why I built ParallelOS.
Most AI advice for physicians is not written by physicians. ParallelOS is a physician-led practice that helps individuals and independent clinics build and operate AI safely, with a focus on local models and human review.
The uncomfortable version of the AI-in-medicine story
Physicians are being told to use AI. The advice is usually not written by a physician.
Most of it starts with the assumption that you will paste your work into a public model, trust an enterprise vendor with your data, and let a black box do things you cannot easily audit.
That is not a serious answer for a working clinician or an independent practice.
What I actually see day to day
I am a practicing radiation oncologist. I run a clinic, a consulting business, trials, and a family. I use AI every day.
The pattern I keep seeing is that thoughtful physicians are stuck between two bad options. Either they avoid AI and lose the leverage. Or they use it in ways that quietly put PHI, consent, and their own judgment at risk.
There is a middle path, but almost nobody is teaching it.
Local models first
The single most important decision in medical AI is where the model runs.
A model that runs on hardware you control changes the risk profile completely. Data stays where you want it. You do not have to trust a vendor's future terms of service. You are not building your practice on a platform someone else can change.
Local models are not the answer for everything, and they are not always the easiest path. But they are the safe default for medicine, and they are getting better fast. ParallelOS is built around that default.
Safety and humans in the loop
The second decision is where a human has to stay in the loop.
AI can draft, summarize, retrieve, and organize. It should not be making consequential decisions on its own inside a clinic. Every ParallelOS engagement has explicit review checkpoints for anything that matters.
I am also careful about scope. ParallelOS is not clinical decision support. It is an operational and administrative layer that lets clinicians do their real work with less friction and more clarity.
Right-sized for individuals and independent practices
Enterprise consultants build for large systems. That is not who I serve.
ParallelOS is right-sized for two audiences. Individual high-complexity professionals who want AI working for them in daily practice. And independent clinics, especially small groups, that want to add AI to operational or administrative workflows without becoming a vendor case study.
Both start with the same spine. Design for safety. Build with local models where they fit. Operate with humans in the loop.
Education, lectures, and the newsletter
Consulting only reaches a small number of people. Most physicians need something else.
So I am building three other paths. A physician-facing education track that starts with the basics and moves into working topics like prompting, workflow design, and safe local deployments. In-person best-practice lectures for clinics and facilities, which are being prepared for CME accreditation through a partner. And a weekly newsletter for physicians about what is real and what is hype in AI in medicine.
The newsletter is the easiest way to follow along. It is called Practical AI for Physicians and lives on Beehiiv.
What I am not building
There are hard boundaries.
ParallelOS is not a replacement for legal, financial, medical, tax, or other professional advice.
It is not clinical decision support.
It is not an autonomous actor that takes irreversible actions without approval.
It is not a way to train on private client or patient data.
The goal is narrower and more useful. Help individuals and independent practices build and operate AI safely, with a system they own and a human kept in the loop.
The invitation
If you are a physician trying to figure out how to use AI without giving up your data, your judgment, or your sanity, this is for you.
If you run an independent practice and you want a physician-led approach to AI safety rather than a vendor pitch, this is for you.
Start with the newsletter, and reach out when a specific need lines up with what I do.
The right question is not whether to use AI. It is where the model runs, whose data it sees, and who stays in the loop.
That is what ParallelOS is here to answer.