How I Used a Local AI to Help My Mother

My mother has cancer. I used a local AI model to read her oncology records in Hebrew, cross-reference treatment protocols, and ask the questions her doctors didn't have time for.

My mother sent me her oncology report as a photo of a photo of a printout. The image was slightly tilted. The text was in Hebrew. The margins were full of handwritten notes from a doctor whose handwriting looked like seismic data.

I am a paramedic. I have spent twenty years fighting for patients to be heard. I have testified to parliaments, trained thousands of professionals, and built entire organizations around the principle that patients are experts in their own bodies.

And I sat there at my desk in Vienna, looking at a blurry photo of my mother's cancer records, and I didn't know what to do.

Here's what they don't tell you about being a patient advocate.

The day it's your family, all the frameworks fall apart.

I know how to teach someone else to advocate. I know the scripts. I know which questions to ask and in what order and with what tone of voice. I wrote a book about it. I've given a talk about it in twenty-seven countries.

And then it's your mother. And suddenly you're not an expert. You're a son. And being a son is much harder than being an expert.

The oncology report was real. The cancer is real. I won't write about the specifics here — that's her story, not mine. But I will tell you what I did next, because I think it matters.

I opened a local AI model on my laptop. Not ChatGPT. Not anything that sends her records to a server somewhere. A model running entirely on my machine. Her data never leaves my desk.

I fed it the report. I asked it to translate the Hebrew into plain English — not medical English, plain English. Then I asked it to explain each finding in the context of her specific diagnosis. Then I asked it to cross-reference the proposed treatment protocol with current clinical guidelines.

It took eleven minutes.

In eleven minutes I had:

A plain-language translation of a document that had been incomprehensible to me an hour earlier.

A comparison of her proposed treatment against four recent studies.

A list of questions I hadn't thought to ask.

One flag — a drug interaction that the report didn't address — that I then brought to her oncologist at the next appointment.

The oncologist confirmed the interaction was worth noting. He adjusted the protocol.

I want to be careful here.

The AI did not diagnose my mother. The AI did not treat my mother. The AI did not hold my hand while I read those records at 2am with a cup of cold tea getting colder on my desk.

The AI gave me information I didn't have. It gave me language to work with. It gave me questions where I had only fear.

Machines handle information. Humans handle transformation.

What I did with that information — how I talked to my mother, how I sat with her, how I helped her feel less alone in a system that had suddenly started using words she didn't recognize — that part was mine. And hers. And there is no model that does that.

I have spent years arguing that AI in healthcare is not a replacement for the human relationship. I still believe that. I believe it more now, not less.

But I also spent years arguing that patients — and the people who love patients — deserve access to the same information the system has about them.

That's the gap I was sitting in at 2am. My mother's records were *about* her. They belonged to her. And yet the language, the format, the density of clinical shorthand — all of it created a wall between her and her own body.

What I used the AI for was translation. Not translation from Hebrew to English. Translation from the system's language to ours.

This is not an advertisement for AI. It's not a technology story.

It's a story about information asymmetry. About the gap between what the system knows and what it communicates. About what happens when a patient — or a patient's son — is handed a document full of facts about a body they love, and the document is written for someone else.

I've been that patient. I spent ten years in bed before I was diagnosed with POTS. I learned to read medical literature from scratch because no doctor would talk to me in a language I could understand. I eventually diagnosed myself. (I know how that sounds.)

What I used then was PubMed, late-night Wikipedia spirals, and a clinical friend who'd answer texts at midnight.

What I used for my mother was a local model that ran on my laptop and asked for nothing in return.

The tools get better. The gap they're filling doesn't go away. The gap is structural.

Here's what I want to say to the person reading this who has a parent with a diagnosis they don't understand.

You are not stupid. The report wasn't written for you.

You are not wrong to want to understand. Understanding is not overstepping.

And if you use a tool — any tool — to help you ask better questions, that's not cheating. That's participation.

The best outcome of any medical interaction is a patient (or a patient's family) who understands what's happening and can engage with the decisions that are being made about their body.

Anything that gets you closer to that is not replacing the doctor. It's doing the work the system didn't have time to do.

My mother is doing okay. That sentence costs me something to write. The okay is real and the uncertainty underneath it is real and both of those things are true at the same time.

Her oncologist is good. The system, in her case, is working — which is not nothing.

And somewhere in the margin of her original report, in the doctor's seismic handwriting, there was a note that I still haven't fully decoded.

I'll bring the AI back for that part too.

I write about making the invisible visible — in healthcare systems, in organizations, and in the lives of people navigating both. If something here landed, I'd like to know.*