ja中文 contact

00technology / measured 2026.10.03

Technology

buildpassingframeworknoneanalyticsga4 onlyfontsself-hostedapi/api/v1a11yreduced-motion

We build our own software and change it every day. Every number and chart on this page is measured directly from our internal repository.

changes to our software

6,500+

Changes merged into our software since 2026.03.20. Each one passed review before going into production.

2026.03.20 — 2026.10.03

fig.1 changes / monthChanges per month
05001,0001,5002,0002,500maraprmayjunjulaugsep
fig.2 every day since 2026.03.20 — 1 cell = 1 dayThe days we changed our software (1 cell = 1 day)
monwedfrisun2026.03.202026.03.212026.03.222026.03.232026.03.242026.03.252026.03.262026.03.272026.03.282026.03.292026.03.302026.03.312026.04.012026.04.022026.04.032026.04.042026.04.05apr2026.04.062026.04.072026.04.082026.04.092026.04.102026.04.112026.04.122026.04.132026.04.142026.04.152026.04.162026.04.172026.04.182026.04.192026.04.202026.04.212026.04.222026.04.232026.04.242026.04.252026.04.262026.04.272026.04.282026.04.292026.04.302026.05.012026.05.022026.05.03may2026.05.042026.05.052026.05.062026.05.072026.05.082026.05.092026.05.102026.05.112026.05.122026.05.132026.05.142026.05.152026.05.162026.05.172026.05.182026.05.192026.05.202026.05.212026.05.222026.05.232026.05.242026.05.252026.05.262026.05.272026.05.282026.05.292026.05.302026.05.31jun2026.06.012026.06.022026.06.032026.06.042026.06.052026.06.062026.06.072026.06.082026.06.092026.06.102026.06.112026.06.122026.06.132026.06.142026.06.152026.06.162026.06.172026.06.182026.06.192026.06.202026.06.212026.06.222026.06.232026.06.242026.06.252026.06.262026.06.272026.06.282026.06.292026.06.302026.07.012026.07.022026.07.032026.07.042026.07.05jul2026.07.062026.07.072026.07.082026.07.092026.07.102026.07.112026.07.122026.07.132026.07.142026.07.152026.07.162026.07.172026.07.182026.07.192026.07.202026.07.212026.07.222026.07.232026.07.242026.07.252026.07.262026.07.272026.07.282026.07.292026.07.302026.07.312026.08.012026.08.02aug2026.08.032026.08.042026.08.052026.08.062026.08.072026.08.082026.08.092026.08.102026.08.112026.08.122026.08.132026.08.142026.08.152026.08.162026.08.172026.08.182026.08.192026.08.202026.08.212026.08.222026.08.232026.08.242026.08.252026.08.262026.08.272026.08.282026.08.292026.08.302026.08.312026.09.012026.09.022026.09.032026.09.042026.09.052026.09.06sep2026.09.072026.09.082026.09.092026.09.102026.09.112026.09.122026.09.132026.09.142026.09.152026.09.162026.09.172026.09.182026.09.192026.09.202026.09.212026.09.222026.09.232026.09.242026.09.252026.09.262026.09.272026.09.282026.09.292026.09.302026.10.012026.10.02

as of 2026.10.03 · source: internal repository · automated changes excluded

fig.3 latest merges — number, date, kindLatest merged changes (titles withheld because they include internal details)

~/clinic-os (main) $ git log --merges -n 40

  1. #12610docs2026-10-03merged into main
  2. #12609feat2026-10-03merged into main
  3. #12607fix2026-10-03merged into main
  4. #12606docs2026-10-03merged into main
  5. #12604chore2026-10-03merged into main
  6. #12602docs2026-10-03merged into main
  7. #12601feat2026-10-03merged into main
  8. #12600fix2026-10-03merged into main
  9. #12595docs2026-10-03merged into main
  10. #12593fix2026-10-03merged into main
  11. #12591chore2026-10-03merged into main
  12. #12589feat2026-10-03merged into main
  13. #12588chore2026-10-03merged into main
  14. #12587docs2026-10-03merged into main
  15. #12583feat2026-10-03merged into main
  16. #12581feat2026-10-03merged into main
  17. #12577feat2026-10-03merged into main
  18. #12575feat2026-10-03merged into main
  19. #12574feat2026-10-03merged into main
  20. #12573feat2026-10-03merged into main
  21. #12571feat2026-10-03merged into main
  22. #12568feat2026-10-03merged into main
  23. #12567feat2026-10-03merged into main
  24. #12565feat2026-10-03merged into main
  25. #12564feat2026-10-03merged into main
  26. #12561fix2026-10-03merged into main
  27. #12560feat2026-10-03merged into main
  28. #12556feat2026-10-03merged into main
  29. #12555feat2026-10-03merged into main
  30. #12554feat2026-10-03merged into main
  31. #12553chore2026-10-03merged into main
  32. #12550docs2026-10-03merged into main
  33. #12546fix2026-10-03merged into main
  34. #12543docs2026-10-03merged into main
  35. #12542fix2026-10-03merged into main
  36. #12541feat2026-10-03merged into main
  37. #12539fix2026-10-03merged into main
  38. #12537feat2026-10-03merged into main
  39. #12536chore2026-10-03merged into main
  40. #12534docs2026-10-03merged into main

01process

With AI, the people who build and the people who use become the same team.

Until now, the companies that built healthcare software and the medical institutions that used it were separate. Builders had to prioritize features common to many institutions, and users who asked for specific changes often waited a long time to see them.

AI has sharply reduced the cost and time of building software, so the people who know the front line can now build systems around how they actually work. At PreMed, every employee is an FDE (Forward Deployed Engineer). We develop as a small team working with AI agents, and what we notice when test-booking on a smartphone, or what front-line staff say after looking at a screen, usually reaches the staging environment the same day.

fig.4 from feedback to productionHow feedback becomes part of the system
01Feedbackfeedback02Changepull request03Checks + reviewchecks + review04Stagingstaging05Productionproduction

02one os, not point solutions

Some things point-solution SaaS cannot reach.

Booking, intake questionnaires, medical records, patient messaging, acquisition metrics — the work of a medical facility is connected, yet most systems so far have covered only one slice of it. PreMed builds all of these as a single OS, in the same team as the front line. clinic os, built for clinics, runs in the group's clinics, and hospital os, built for hospitals, is in development.

fig.5 point solutions vs. one osPoint-solution SaaS and a single OS
point solutionsPoint solutionsMarketingsystem ABookingsystem BIntakesystem CClaimssystem DAnalyticssystem Epeople copy by handcopy & paste between systemsrequests wait months for a releasefeature requests wait for the roadmapnumbers live on separate screensdata stays in silosone osOne os (clinic os)The clinic floor (staff · patients)the floor — where we work every dayfeedback ships the same dayMarketingBookingIntakeClaimsone data model · one set of numbersbuilders and users are one team · updated dailybuilders and users are the same team
point solutionsPoint solutionsMarketingsys.aBookingsys.bIntakesys.cClaimssys.dAnalyticssys.epeople copy by handmanual copy between systemsrequests wait months for a releaserequests wait for a roadmapnumbers live on separate screensdata stays in silosone osOne os (clinic os)The clinic floor (staff · patients)the floorfeedback ships the same dayfeedback → release, same dayclinic osMarketingBookingIntakeClaimsone data model · one set of numbersbuilders and users are one team · updated dailybuilders and users are the same team
  1. 01

    No manual work left at the seams

    When systems are separate, copying and checking data between them remains a human job. With a single OS, the seams themselves disappear.

  2. 02

    Fitted to our operations, not to the average

    Products built for many institutions have to prioritize common features. We build around how the group's clinics work, and can fix things the same day.

  3. 03

    One continuous view of the numbers

    From the first point of patient acquisition through booking to the visit itself, everything sits on the same data and can be followed on a single screen.

  4. 04

    The more clinics, the better the OS

    Clinics that join through succession and newly opened clinics all run on the same OS. An improvement found at one clinic reaches every clinic.

  5. 05

    Updated daily, and embedded on the front line

    The front line changes every day. Updating once every few months is, in our view, already too slow. We update our software every day (6,500+ changes since 2026.03.20). What matters is building an organization that can embed each change on the front line, and the execution to see it through.

03galapagos

Galapagos, our booking system

An AI-native booking system for medical facilities, built in-house by a small team working with AI agents — people who know clinical practice first-hand. It goes into use at Kids Clinic Marumaru, opening in November 2026, and will then be extended to the group's other clinics, including gynecology and fertility care.

Play the Galapagos film ("Rethink booking.")
video galapagos — brand filmFilm on YouTube (in Japanese). Playing it connects to YouTube
  1. 01

    Patients book entirely within a LINE chat

    Booking, rescheduling, cancelling, joining a waitlist, adding family members and filling in the questionnaire can all be done without leaving LINE. Patients can tap through buttons, or simply write “Tomorrow morning” or “I’d like to cancel Friday’s appointment.” When a patient makes a request in free text, AI checks availability and suggests options; a booking is confirmed only when the patient presses the button themselves.

  2. 02

    AI handles first response; only unanswered inquiries reach staff

    AI answers questions sent through LINE within the information the clinic has prepared. Only what AI could not answer, or what needs a staff decision, reaches the admin screen — and once staff add an answer, AI can answer it next time. It never answers symptom questions on its own judgment, and when it detects urgent wording it directs the patient to call 119 or the clinic.

  3. 03

    AI agents carry out detailed admin settings

    Most settings — opening hours, the number of booking slots, services, questionnaire items, notices to patients — can be changed by asking an AI agent, without opening the admin screen. A questionnaire of over 100 questions can be registered at once from a table. What an AI agent can do is governed by role-based permissions, and every change is logged with who made it.

  4. 04

    Bookings and questionnaires sync automatically with the claims system and EMR

    New patients are registered in the claims system as soon as their booking is confirmed, so their patient ID is set before the visit. Reception actions are reflected in the claims system too, and questionnaire answers flow automatically into the electronic medical record. Reception and consultation rooms no longer need to copy the same information by hand.

  5. 05

    Automated messages before and after each appointment

    Day-before reminders, waitlist promotions and return-visit messages for patients who haven’t come in for a while are sent automatically. Each message checks that the booking is still valid just before sending, so people who have cancelled never receive a reminder the day before. For families, every registered guardian receives it.

  6. 06

    Works for any specialty, with features added for each

    Services, consultation durations, questionnaires and notices to patients are set freely by each clinic. Features are added to fit the way each specialty works, and clinics can switch off the ones they don't use. Pediatrics, where it goes into use first, is one example: services shown by the child's age in months or years, last-minute slots for sudden symptoms, combined bookings for siblings or parent and child, vaccine stock linked to bookings, vaccination records entered by parents, age-based notices, fever charts and growth curves.

  7. 07

    Acquisition tracked all the way to the visit

    For each entry point — ads, website and others — it records how far each patient got through booking and whether they actually visited, in one continuous record. For LINE bookings it also shows where people dropped out and how many could not book because slots were full, so “bring in more patients, or add booking slots?” can be decided on the numbers.

fig.6 how a message flowsHow inquiries and bookings flow
Patient (LINE chat)patient on lineinAI first responseai first responseChecks slots, suggests timessuggests open slotsPatient presses to confirmpatient confirmsBillingbillingEMRemrSynced automatically, no re-typing↑ only when AI cannot answerStaff inboxstaff inboxDirects to 119 or the clinicurgent wordsAI never judges symptoms itselfDay-before reminderremindersre-checks before sending
Screens in a LINE chat from choosing a time to a confirmed booking
01Booking completed entirely in a LINE chat
Admin screen listing only the inquiries AI could not answer
02Staff see only the inquiries AI could not answer
Screen where an AI agent is asked to close the afternoon session
03Booking slots and questionnaires can be changed by asking an AI agent
Fever chart drawn from temperatures entered in the questionnaire
06Temperatures entered in the questionnaire become a fever chart
Growth curve drawn from height and weight records
06Growth curves drawn from height and weight records
Screen showing patient numbers continuously from ads and website through LINE and booking to the visit
07One continuous view from ad to visit

* Screens show fictitious demo data (images from the press release).

Press release (PR TIMES, 7 October 2026, Japanese) →

04toki — fertility care database

TOKI, an EMR for fertility care

The electronic medical record used every day in the group's fertility care. It brings together what fertility care needs, from treatment cycles to IVF and frozen embryos, on screens each doctor finds easy to read. When something is missing, we build it quickly from the floor's requests.

  1. 01

    One set of data, a screen for each doctor

    The same patient data appears in a different UI for each doctor. A doctor who follows the course of treatment line by line gets a single continuous view; a doctor who compares many items at once gets three panes side by side. Everything has its fixed place, and what is needed fits on one screen.

  2. 02

    Every record fertility care needs

    Treatment cycles and daily visits, IVF, frozen embryo transfer, semen analysis and self-injections are all handled in one database. Things that are easy to miss, such as embryo storage deadlines and uncollected consent forms, show up in a list.

  3. 03

    Today's work on one screen

    Today's outpatients, today's lab, today's self-injections: the patients to see and the tasks to do each fit on a single screen. In the lab, double checks by two people are recorded too.

  4. 04

    Easy to read, easy to enter

    Insured and self-pay care are color-coded so they can be told apart at a glance. Small chores in data entry are taken away, and half-written entries are never lost.

  5. 05

    Requests become features quickly

    A button at the bottom right of the screen lets users send "it would be nice if…" on the spot. Each request is registered directly as a development issue, and we build it with AI agents. A request sent in the morning has been implemented by early afternoon the same day. In the roughly five weeks since late August 2026, more than 170 changes have gone into TOKI.

fig.7 one data, many layoutsOne set of data, a screen for each doctor
Patient data (one model)one data modeldataStandard viewstandard viewPer-doctor viewper-doctor viewrequests from the floorSend a requestsend a requestBecomes an issuebecomes an issueShipped to the screenshipped to the screenA morning request can be on screen the same day
Standard screen showing a patient's course of treatment in a single column
01The standard screen: the course of treatment at a glance
Doctor's screen arranging the same patient data densely across three panes
02The same data across three panes, all on one screen

* Screens show fictitious demo data.