A second brain for a medical emergency

A family member had a medical emergency recently. They’re home now. This post is not about them — I’m keeping the who, the where and the what deliberately vague — it’s about the thing I discovered in the middle of it, which is that a hospitalisation is also an information disaster, and almost nobody is equipped to handle that part.

Here is what arrives at you, in about thirty-six hours:

• A discharge packet, photographed on a phone, in the form of image scans with no text layer.
• A medication list with doses and intervals, none of which you recognise.
• Four follow-up appointments with four different specialists, none booked.
• A stream of people in scrubs who introduce themselves once, quickly, while you are not at your best.
• A diagnosis that is not the same as the one you were given on admission.

You are expected to absorb all of it while worried, sleep-deprived, and in a building you’ve never been in before. Then you’re expected to still know it in three weeks when a doctor asks what the discharging physician said.

I couldn’t. So I built somewhere to put it.

Everything goes in one place, raw

The first rule was that nothing gets summarised on the way in. Phone photos of paperwork, portal exports, screenshots, my own notes — all of it lands raw and stays raw, and the distilled version lives separately. If the interpretation turns out to be wrong later, the source is still there to re-read.

The scans were the immediate problem: images, no selectable text. They got rendered page by page and read as images. That worked far better than I expected — including the handwriting, the checkbox forms, and the medication table that mattered most.

The wearable turned out to be evidence

The part I did not anticipate: a consumer smartwatch had been quietly recording something relevant for months before the emergency, and had flagged it more than once.

The hospital’s chart described the finding as new. The watch history said otherwise, plainly, with dates. That is not a small distinction — it changes how long something has been going on, which changes what the treatment plan should be. Pulling that export and putting it next to the clinical record turned a vague worry into a specific, dated, documentable thing to hand a specialist.

It’s an odd feeling to realise the most useful longitudinal data in the room came off a wrist rather than out of a hospital system.

Who was that doctor?

Over a short stay a genuinely surprising number of clinicians pass through. They say a name and a specialty and then they’re gone, and later you’re trying to recall whether the person who said the reassuring thing was the cardiologist or the hospitalist.

So the system keeps a roster: every clinician who appeared, their specialty, what they said, when, and — where I could find one — a photograph, so I could actually match a face to a name to a piece of advice. Alongside that, licence verification: are they board certified, are they in good standing, what does the public record say. Not out of suspicion. Out of wanting to know who I was talking to when I was too rattled to take it in properly at the time.

The hallway conversations

I started recording conversations on my phone. Transcribed, they became searchable alongside everything else — and that’s where the system started earning its keep, because it could put a transcript next to a chart and notice they disagreed.

Not dramatically. Just the ordinary drift of a big institution: one document saying a thing is new, another implying it isn’t; a medication on the discharge list that sits oddly next to another medication on the discharge list; a follow-up described as booked that had no date attached to it anywhere.

Three or four of those surfaced. Each one became a question.

What it actually gave me

A question list. Not vague anxieties — specific, sourced questions with the contradiction attached, ready to walk into an appointment with. This is the single highest-value output of the whole thing.

Calendar entries. Four unbooked follow-ups with different deadlines became actual reminders, and the medication supply had an end date that became its own alert. It is startlingly easy to lose an appointment window in a fortnight of worry.

Permission to ask dumb questions. This is the part I want to emphasise, because it’s the one I didn’t see coming. I could ask what a term meant, then ask again because the first answer went over my head, then ask a third follow-up, at two in the morning, with no one sighing at me and no clock running. Anyone who has tried to get a fourth clarifying question out of a consultant with a ward to finish knows exactly how valuable that is.

Charts and diagrams on demand. Timelines, medication schedules, a diagram of what was actually going on. Whatever shape helped, in about a minute.

A record that outlives the crisis. If there’s a next time, none of this starts from zero. There’s a complete, dated, sourced history to hand a new doctor, instead of a family member trying to remember what happened last year.

And then there was the travelling

The emergency also meant getting on a plane at short notice for a few days, which is its own small logistics problem layered on top of a bad day.

The constraint that turned out to govern everything wasn’t price — it was visiting hours. Once you know the ward closes at a fixed time in the evening, most of the cheap same-day options evaporate, because arriving at 10pm buys you nothing but a hotel room. Working the flight options backwards from “how many hours does this actually put me at the bedside” produced a very different answer than sorting by price would have.

It also produced one genuinely useful piece of arithmetic. Flying out that same night versus first thing the next morning was a difference of roughly a thousand dollars, all in — and because of the visiting cutoff, it bought about forty extra minutes at the bedside. Seeing that stated plainly made the decision easy in a moment when I was in no state to do that maths myself.

Everything else stacked on top of the same idea: rank the options by the thing that actually matters, not the thing that’s easiest to sort by.

• Nonstops only, and realistic door-to-bedside times rather than landing times — deplaning, the rental shuttle, the counter queue, the drive at that hour of day.
• Hotels ranked by drive time to the hospital, not by star rating or price. The winner was a few miles closer on the drive that happened twice a day, and slightly further from the airport on the drive that happened twice a trip. That’s the right way round, and it isn’t obvious from a map.
• Car rental at the airport, never an off-site branch, because at the end of a long day the shuttle is the thing that breaks you.
• Somewhere to eat, which sounds trivial until it’s 9pm in an unfamiliar town and you’ve not eaten since the airport.

Then, afterwards, the money. Every charge from the trip pulled together and reconciled against what had been booked — and two things immediately looked wrong. The rental car had been authorised for a couple of hundred dollars more than the contracted rate, and the hotel for a round hundred over the booking. Both turned out to be ordinary pre-authorisation holds rather than real charges, but I only knew that because the actual receipts were sitting next to the card transactions where the difference was obvious. One of them also revealed I’d have been billed for a night I didn’t stay, if I hadn’t checked.

Estimated cost versus actual cost, itemised, with the gaps explained. It’s a small thing next to the medical side. But it’s one more thing I didn’t have to hold in my head, at a time when I had no spare capacity for any of it.

The part that actually mattered: it already knew us

Everything above makes it sound like I built a medical tool in a hurry. I didn’t. Almost none of this was built for the emergency. What existed was a second brain that had been accumulating for a long time — a growing corpus of context about me and my family, kept as plain files, added to a bit at a time whenever something was worth writing down.

The emergency didn’t need a new system. It just drew on that one.

Which is why the help it gave was contextually aware in a way a fresh chat window can never be. I never had to explain any of the following, because it already knew:

• That I fly nonstop, and which carrier I’d rather be on.
• That I want the rental car picked up in the terminal, never an off-site branch — a preference I’d formed after a previous trip went badly, and had written down precisely so I’d never have to re-litigate it.
• Where my family stays in various places, and where they don’t, so no one wasted time booking lodging that wasn’t needed.
• What we like to eat, and what we avoid.
• Who everyone is, how they’re related, and who needed to be told what.
• What was already on my calendar that would collide with being away.

None of those are impressive facts individually. Collectively they’re the difference between a tool you have to brief and an assistant that already has the brief. Every one of them was a decision I didn’t have to make, or explain, or make badly at eleven at night.

And that compounds in the other direction too. Because the corpus was already there, the emergency’s own material — the scans, the recordings, the clinician roster, the trip and its costs — didn’t land in a vacuum. It landed next to everything else, cross-referenced against it, and became part of the same searchable body of knowledge. The wearable history only became evidence because years of it were already sitting there. The travel decisions were only fast because the preferences were already written down.

The model is not the valuable part. Models are interchangeable and they get better every few months regardless of what I do. The valuable part is the corpus: a large, curated, well-sourced body of context about one specific family, that I can ask anything of, in plain English, and get an answer that accounts for all of it at once.

That’s what a second brain actually is. Not a chatbot with a good memory — a place where the accumulated detail of a life is written down in enough order that it can be reasoned over, by me or by something helping me, at the exact moment I’m least capable of holding it in my own head.

I’ve been building it casually for a while, mostly as a hobby. It stopped being a hobby the week I needed it.

What I’d tell someone about to do this

• Start before you need it. The corpus is the whole asset, and it cannot be assembled during a crisis. Everything that helped me that week was written down on some ordinary afternoon months earlier, by someone who had no idea what it would eventually be for.
• Capture first, understand later. In the moment you cannot tell what will matter. Photograph everything. Sort it out afterwards.
• Never overwrite a source. The wrong early interpretation is evidence about how you got confused. Keep it, mark it superseded, move on.
• Contradictions are the product. The value isn’t the summary — a summary smooths over exactly the bumps you need to see. The value is “these two documents disagree, here’s the question that resolves it.”
• It is not a doctor and must not act like one. Nothing here diagnosed anything or changed any treatment. It organised, cross-referenced, and helped me ask better questions of the actual physicians. That boundary matters, and the moment it blurs, this stops being useful and starts being dangerous.

The technology involved is not exotic. Image reading, transcription, a calendar API, a pile of markdown files and a lot of care about provenance. Any of it could be rebuilt in a weekend.

What can’t be rebuilt in a weekend is the context. Two years of writing things down — preferences, relationships, decisions and why they were made, the boring administrative substrate of a family — turned out to be the thing that made all of it work when it counted. Not because any single entry was important. Because together they meant I could ask one question and get an answer that already accounted for everything.

I was frightened, exhausted, and a long way from home, and I had somewhere to put it all and something to ask. That’s the thing I’d want anyone in the same position to know is possible — and the reason to start building it now, on an ordinary afternoon, while nothing is wrong.