Collection
Healthcare

The brief, as it was actually written: clinics of different sizes work in different ways, and consultation time is the thing to cut. 
Three segments, each asking for something different. The largest asked for AI by name, for resource management and analytics, which is why there is a model in the architecture three frames down. - Small, all manual

- Medium, partly digital

- Large, admin over departments

The same clinic drawn at three sizes. Small runs on one doctor, two nurses and a reception desk. Large grows an admin above departments that each repeat the whole structure. Any role model that fits one of these fits neither of the others. 
Where the AI actually sits: it returns three possible diagnoses and a plan, and two human validations gate it. Symptoms are confirmed by the patient before the model sees them, and the plan is confirmed by the doctor before it moves forward. - Language first

- Their own words

- Follow-ups

- They confirm it

The patient triages themselves before arriving. Language first, then their own concerns, then follow-ups per concern. The last frame is the gate: they read back exactly what will reach the clinician, and confirm it. 
The queue as the patient sees it: how long, where they are in it, and what the visit will consist of. One mechanism, first of three seats. 
The same queue as the nurse works it: length, average wait, throughput per hour, a hold control, and every patient one click from no-show, reschedule or record vitals. 
And the same queue as the doctor composes it: a whole day as one board, who is registered, who is waiting, and where consultation time actually goes. Three seats, one mechanism, designed three times. 
Every panel is a widget, so a doctor composes the board they need instead of the one the product assumed. 
The chooser behind that board. Six block types, and the doctor decides which of them their day needs. One of those widgets, rebuilt to run: idle, editing in place, running, paused, and the task row's two separate exits. Click to interact. - Idle

- Editing in place

- Running, End timer

- Task attached

- Task ticked

- Task dismissed

- Break Timer

- Focused Consultation

The same timer under four titles, across idle, editing, running and done. One primitive doing four jobs, which is what a widget has to be if the board is composable. - How long, and why

- On hold, with a way out

- The nudge back

Holding the queue, end to end: how long and why, then the hold itself with a way out of it, then the nudge back. A doctor stepping away is a state the whole hospital can see. 
Inside the consultation, symptoms and diagnoses arrive as suggestions the doctor selects from, and each one gets qualified: to rule out, suspected, or follow up. 
Then the model returns three candidates and says how sure it is about each: 80, 50 and 30 percent. It does not decide. Treatment stays locked until a doctor picks one. 
Then it recommends medicines, and the doctor keeps, drops or deletes them one by one. The preview composes the instruction in plain English, so what the patient will read is checked before it is a prescription. - How the practice is running

- Fact bolded, inference hedged

What the clinic learns from all of it. Peak hours, new against returning, top diagnoses and medications, and an insights panel that bolds what the data says and hedges what it only suggests. - Accept or decline

- As a pipeline

A referral is not a message, it is a state: received, contacting the patient, scheduled, completed, follow-up required. Accepting one is a decision with the clinical detail attached. 
One record, written by several products, and it says so. Each block names which one last touched it and when, so a nurse can tell where a fact came from before trusting it. 
And who is allowed into any of it: QMS, NMS and Labs as products an organisation buys, with User, Admin or View Only granted per person per product. 
The whole estate, and the reason the queue board could not be designed alone: check-in, QMS, patient management, lab, and the surfaces each role actually touches. - The day, in numbers

- Provisional, then final

- Examination as a checklist

- Referrals

The doctor's half of that estate in a pocket: the day by patient stage and ward, then a patient profile that runs provisional diagnosis into a structured examination, inspection through percussion, ticked off as findings. - Home

- Virtual or walk-in

- What happens next

- The family's records too

And the patient's half: appointments virtual or walk-in, what happens after the call, and a profile that holds the family's records beside your own. - Doctor App, home

- The hospital screen

- Reports, unaccompanied

- Logging a condition

The generation before those two: a doctor's home, and the patient side of the same appointment with reports they can open unaccompanied and a record they keep between visits. - Splash

- Login

- Home

- Switch proxy

MyMed, the same patient problem redrawn for a different market as a trial. Proxy switching is the part worth looking at: one person holding their own record and their family's, which is how care actually gets managed. 
The same permission moments drawn twice, iOS on the left and Android on the right, rather than once and hoped for. - Delivery status

- Identify a pill

- Which profession

- Sign in

DaxP, a freelance healthcare app: delivery status stated plainly enough to stop the follow-up call, a pill identified by pointing a camera at it, and two professions separated before the first screen rather than after it.