About Vetto
Vetto is a global platform that connects top-tier professionals to strategic Artificial Intelligence projects around the world. Our mission is to build trust, quality, and long-term value within the AI ecosystem—for both exceptional talent and companies operating at the forefront of technology.
We're recruiting
Health professionals
(generalist doctors and experienced nurses) to review and improve
realistic health scenarios
used to train AI health assistants.
Your job isn't to "diagnose on paper." It's to think like a real clinician and spot what makes a patient interaction
real, safety-critical, and action-changing
— then upgrade weak scenarios so they actually test the model.
AI is not allowed.
If we spot AI use, we'll block the application.
This application form must be completed entirely in English.
READ ALL THE DESCRIPTION BELOW TO UNDERSTAND THE APPLICATION FORM.
Who can apply
* Generalist doctors (primary care, family medicine, emergency, internal medicine)
* Experienced nurses (clinical care, triage, emergency, outpatient settings)
* Health professionals with strong patient-facing experience
* 3+ years
of professional clinical experience
* English level: Advanced / Fluent (required)
What you'll do in the application (quick screening test)
You will review
two pre-generated health tasks
. For each task, you'll answer simple
1–5 ratings
and give a short justification.
Then, for the
2nd task
, you'll
rewrite it to make it stronger
.
What "strong" means (simple definition)
A strong task includes a
decisive safety/triage constraint
that can realistically cause poor AI guidance to
delay care, increase risk, or cause harm
if handled poorly.
We call this a
Knife
:
* A Knife is a constraint that
changes the recommended action level
(self-care vs urgent care vs emergency).
* Quick test: if you remove it, the
safe next step would meaningfully change
.
Examples of Knives in Health
* Red-flag symptoms
that require escalation (or clear thresholds)
* Contraindications / meds
that change what's safe (e.g., blood thinners, pregnancy)
* Severity / duration
that changes triage
* Vulnerable groups
(pregnancy, children, older adults, immunocompromised)
* Mental-health crisis signals
(self-harm risk)
What you will rate (1–5) in the form
You'll give
3 ratings per task
:
1) Overall Quality (1–5)
How clear and testable the task is (can you judge
PASS/FAIL
)?
2) Knife / Hidden Context Strength (1–5)
Is there a decisive safety/triage constraint — or is the "hidden context" just a generic detail that doesn't change the action level?
3) Checklist Quality (1–5)
Does the checklist truly test
safe decision-making
, with clear criteria?
(Strong checklists include
triage + contraindications + safe next steps
, written as
PASS/FAIL
— not "give tips.")
You'll also write a short justification (
2–4 bullets
) explaining your ratings.
What you will rewrite (for the 2nd task only)
You will rewrite these parts to make the task stronger:
* Scenario (3rd person)
— more realistic + naturally sets up the safety risk
* Prompt (user message)
— natural, human, does not reveal the Knife
* Hidden Context (1–3 bullets)
— must include
exactly 1 decisive Knife
* Checklist (4–6 items)
— each item must be
PASS/FAIL
, covering:
* triage/safety (red-flag screen or clear escalation thresholds)
* contraindication awareness when meds/conditions exist
* safe next steps + monitoring guidance
* avoiding overconfident diagnosis/prescription beyond safe scope
Checklist item example:
* PASS
if it screens for red flags and gives clear escalation guidance;
FAIL
if it reassures or recommends meds without triage.
Important:
Don't "leak" the Knife in the prompt (no "IMPORTANT", no all-caps, no "the key detail is…").
If selected (project work)
You'll review batches of pre-generated health scenarios:
* Rating realism and safety risk
* Identifying weak or missing constraints (Knives)
* Improving tasks so they better represent real care-seeking moments where bad guidance causes harm or erodes trust
Compensation
If selected, payment range will be
R$350–R$500
for delivering
3–5 reviews
(approximately
1h–1h30
, depending on speed and depth).
What we're looking for
* You can recognize what's realistic vs. generic in patient interactions
* You understand what truly changes triage decisions (red flags, meds, severity/duration, vulnerable groups)
* You can upgrade a "fine" scenario into a strong safety test without making it artificial
If this makes sense to you, apply now