# Two hours to design triage—and most of the work was defining limits

- Date: 11 May 2026
- Project: sana
- Tags: AI, Health, Hackathon

> A hackathon at Di Tella asked how AI could help where the nearest specialist is six hours away. Our answer was defined less by its features than by the lines it would not cross.

The brief at the Universidad Torcuato Di Tella hackathon stated that some
provinces are a six-hour journey from a specialist and that Latin America has
roughly one doctor for every five thousand people. We had two hours, in teams
of three to five, to design and pitch a response. One constraint was explicit:
**not an appointments app**. The problem began when no doctor was available
and somebody needed guidance now.

![The Sana team presenting its proposal in a classroom at Universidad Torcuato Di Tella](https://eitanf.com/static/sana-pitch-51bb5671.webp 'Presenting Sana at Di Tella, 11 May 2026')

## Our answer: triage, not diagnosis

We proposed **Sana**, a chat that would ask narrowing questions — closer to
Akinator than to a form — using symptoms, age, history and duration. It would
return an orientation: possible causes, a priority level and what to do next.

![A laptop displaying Sana's mobile triage prototype, including its offline and emergency guidance](https://eitanf.com/static/sana-prototype-0e7561bb.webp 'The Sana interface prototype used in the pitch')

## The decisions that were refusals

- **It would not diagnose.** Sana would orient and prioritise. An AI that says
  "you have X" where no doctor can challenge it is worse than no answer.
- **Escalation to a person would not be optional.** Chest pain, trouble
  breathing, fainting, seizures or high uncertainty would lead directly to
  human care or an instruction to seek it immediately.
- **It would not assume a good connection.** Consultations would queue offline
  and sync later, and the system could be used from a community health post
  rather than requiring everybody to have a reliable connection.
- **It would not pretend responsibility belonged to the model.** Clinical
  responsibility would remain with the health system and the professionals
  supervising its protocols, and every response would be labelled as
  preliminary guidance.

## What we built — and what we did not

The team designed and pitched the concept, and Guido Jacofsky and I built most
of the web interface prototype. It made the proposed patient flow tangible,
but it did not make Sana a working clinical system. We did not build or
validate medical triage; we built a prototype that showed how it could feel.

That is where Sana stopped. [The prototype is still online](https://sana-prototype.pages.dev/),
and the project is documented [in the lab](/lab/sana).

## What I take away

With two hours, features are the easy part: any team can describe a chat that
asks questions. The harder and more important work is deciding where it must
stop. In healthcare, the boundaries **are** the product.

---

Page: https://eitanf.com/notes/two-hours-to-design-triage · Español: https://eitanf.com/es/notes/two-hours-to-design-triage.md · llms.txt: https://eitanf.com/llms.txt