The desk and the panel

One queue holds everything patients have asked for. The panel is where the decisions get made — the AI only arrives at them prepared.

Status: Live at the Pinamar installation.

The queue

Every request lands in one list, oldest first, with the AI’s reading of the referral order already attached: what study it asks for, which specialty it belongs to, whether the professional’s signature and licence are there. The person at the desk approves, rejects, or edits and then approves.

Approving does three things at once — it books the appointment, writes it to the health record, and emails the patient. If the health record is down, the first two still happen and the write is retried on its own.

Rejecting picks a reason from a short curated list rather than typing free text, because the reason is what the patient reads, and it is what turns into the instruction for fixing it. The list is edited in the panel when a new reason turns out to be real.

Today’s patients

The day in one screen: who is coming, at what time, to which professional, with what coverage, and whether they answered the reminder. Anyone who arrives without an appointment is added here, so the list matches the waiting room rather than the calendar.

Agendas and the calendar

  • Each professional has an agenda: days, hours, appointment length, which centre, which services.
  • Exceptions are first-class — a holiday, a training day, an afternoon that runs elsewhere. The patient site stops offering those times the moment they are entered.
  • The calendar reflects the real one. What the desk sees and what the patient can take are the same data, not two systems kept in sync by hand.
  • A professional who leaves is disabled, never deleted, whenever they have appointments — the history has to stay readable.

Who sees what

Five roles, and a new endpoint is closed to all of them until it is deliberately opened:

  • Administrator — everything, including users, services and settings.
  • Front desk — the queue, today’s patients, booking on someone’s behalf.
  • Clinician — their own agenda and their own patients, with the pre-consult interview and the record summary.
  • Reports — the numbers, and nothing identifying beyond what a report needs.
  • Communications — the public texts, the news entries and the site’s own copy.

Login is a password plus, optionally, a second factor or a passkey. Sessions are the only source of identity anywhere in the product.

Reports

Each report answers one question and says the answer in a sentence before it draws anything: how much of the offered capacity was actually used, who did not show up and on which services, and how much of the care given to insured patients was never billed back to their insurer. That last one tends to pay for the installation.

Settings and system health

Languages, location, the site’s own texts, the AI instructions and the API keys are edited from the panel, without a redeploy. Keys are shown as present or missing, never printed back. A status screen says which integrations are actually answering right now — the health record, the coverage registry, WhatsApp, email, the AI provider — so “it is not working” has an address before anyone opens a terminal.

Next: Install