Verified clinicians only · Now accepting applications

Your clinical expertise
trains the AI that
helps patients

HoliwellAI is a platform where licensed mental health professionals annotate realistic patient scenarios — building a high-quality dataset that teaches AI to understand human distress.

Apply for access →How it works
Built forPsychiatristsClinical PsychologistsCounsellorsMental Health NursesCrisis Interventionists

Your contribution

Read, respond, label — that's it

Each session presents you with realistic patient statements. You write a clinical response and identify the emotional presentation, risk level, and patient intent — exactly as you would in practice.

01
Read the patient statement
A realistic scenario drawn from anonymised clinical contexts. Ranges from everyday anxiety to crisis presentations.
02
Write your clinical response
Respond as you would in a real session — empathetically, safely, professionally.
03
Label what you observe
Select the emotions present, the risk level, and what the patient is seeking. Multi-label, nuanced, expert.
04
Advance the difficulty level
As you complete more cases you unlock higher-difficulty scenarios — building depth across the dataset.
1
Patient scenario
Intermediate

"I've been feeling really disconnected from everyone lately. Even with my friends I just feel like I'm performing — like there's a version of me watching from outside. It's been like this for weeks."

Emotions observed
lonelinessdisconnectionemptinessanxiety
Risk level
Low
Patient intent
Emotional expression

The process

From application to AI training

1
You apply
Register with your professional licence number and specialization.
2
We verify
Admin team checks your credentials against your country's registration body.
3
Access granted
Once approved, you can begin contributing immediately.
4
You annotate
Complete sessions at your own pace — 15–30 minutes each, any time.
5
AI gets trained
Your annotations feed directly into model training once reviewed.
◎

Expert-only dataset

All annotations come from licensed mental health professionals. No crowdsourcing, no shortcuts. The AI learns from people who understand what a patient is actually expressing.

◈

Safety-first design

Crisis annotations go through a mandatory second review before entering training. We enforce strict safety score thresholds — data that doesn't meet the bar doesn't get used.

◇

Your judgement, preserved

The ontology is built around clinical concepts — not technical labels. You annotate in the language of practice, and that specificity is what makes the model clinically meaningful.

30
clinical presentations
From everyday anxiety to psychosis and identity distress
27
emotion labels
Covering the full spectrum of human distress
8
intent categories
What patients are truly seeking
4.0+
safety score required
Out of 5 — before data enters training

Ready to contribute?

Applications are reviewed within 2–5 business days. Once approved, you can start contributing immediately. There is no minimum commitment.

Apply for access →

Already have an account? Sign in