Emotional Zoo

A meltdown should never be the first sign.

In autistic children, published studies find a signal in the body minutes before an episode. Emozu is testing if a soft band with a little friend on top can warn a parent in time, at home.

Write to us See the evidence

Research stage. Not a medical device. Not for sale.

Concept photo of a ladybug friend on a soft blue band, on a 6-year-old's wrist, real size — concept image, not a photo

99children in all published studies

0of them followed at home

1 to 10 minhow far ahead the models warn

The problem

Today a parent finds out by watching it happen.

This started with one family close to us. Then we read what other parents write, and saw the same picture. The child often cannot say what is wrong. The parent looks for a trigger after the fact, and the next one comes without warning.

47%

of 409 parent posts are about hunting for triggers.

Our tagging of public posts, mostly r/Autism_Parenting.

68%

of autistic children have been aggressive toward a caregiver at some point.

Kanne & Mazurek, 2011

49%

of children tried to run away after age 4.

Anderson et al., Pediatrics, 2012

All numbers and sources

Our idea

A soft band. A little friend on top. Serious sensing inside.

Many children pull a medical sensor off. A friend is different. The friend sits on a band, the band reads the body, and the parent gets a quiet message on the phone. No screen on the child and no alarm in the room.

Exploded render: a ladybug friend with a cavity for the battery, the battery in a dark tray with two hooks, a latch and four contacts, a thin cover with locks and pogo pins, the board with its sensors on the underside, a curved base with holes for them, a soft frame and a woven band

Heart rate and HRV
Skin conductance, EDA
Skin temperature
3-axis motion
Battery in the friend
No screen, ever

In a 2025 survey of 40 parents of nonverbal children, 67.5% supported a wearable stress detector (Weber et al.).

Concept photo of a hedgehog friend on a blue woven band, on a 6-year-old's wrist resting on a table

All day on the wristWoven band, soft frame, nothing to catch

Render of a ladybug friend tipped up on its two hooks above the base, the four spring contacts visible

Off in one moveTwo hooks, a latch, four contacts

Concept photo of a turtle friend on its band on a bedside table in the evening, a parent's phone showing the child's day

At night, nearbyThe parent sees the day in the app

Render

Three friends. One band.

Hedgehog, ladybug, turtle. Soft silicone tops, a soft TPU frame, a woven band. Each friend carries its own battery, so a child can swap friends and keep the band on. At home, friends rest and charge in the Zoo.

Parent app

Warn me, show me the day, tell me what actually helps.

The alert

Minutes before, not after.

Concept screen: Andrew's day in the Emozu app, a calm and tension line with lunch and one alert, 78% calm, 2 alerts, 11 minutes warned early

The day

Patterns nobody can hold in their head.

Concept screen: what helps, ranked by how fast the body settles, friend's slow buzz 8 of 10, leaving the room 7 of 10

What works

Ranked by the body, not by guesswork.

Concept screens with invented data. Nothing here is a working product yet.

What it is based on

The body changes first. That can be measured.

Four studies from 2019 to 2026 use a wearable sensor to catch an episode. Three of them predict it, 1 to 10 minutes ahead. The models look for this: heart rate goes up, heart rate variability goes down, and skin conductance bursts come more often.

The last 20 minutes before an episodeSchematic, not real data

Heart rate ↑ HRV ↓ Skin conductance ↑ warning window

Minutes before the episode. Curves are illustrative. The window comes from Imbiriba 2023 (3 min) and Kartha 2026, preprint (10 min).

3 min

ahead, AUROC 0.80, in 70 young people across 4 hospitals.

Imbiriba et al., JAMA Network Open, 2023

10 min

ahead, AUROC 0.78, in 9 young people in a special-ed class. The sensor sat on the ankle.

Kartha et al., 2026, preprint

of the 99 children in these studies were followed at home.

Hospitals, clinics and classrooms only.

The hard part: about 2 of 3 alerts were false.

A parent who gets pulled away for nothing stops listening. So we do not start with alerts. We start silent: record, label, and look back. “Yesterday at 16:40 it began eight minutes before you noticed” has no false alarms at all. Live alerts come later, and only if the numbers hold.

The gap

99 children in the literature. None of them at home.

Home is where a parent needs the warning most. The research devices the models were trained on left the market: Q-Sensor in 2013, Empatica E4 in 2025.

Goodwin 2019 20

hospital unit

Imbiriba 2023 70

4 psychiatric hospitals

Kartha 2026 9

special-ed class

At home

labeled episodes

0first set: Emozu

One dot, one child. We do not add Bahrami Rad 2025 (11 children): that group likely overlaps with Kartha 2026.

Full research

Roadmap

Where we want to be in 6 to 9 months.

A paper with clinical psychiatrists, on 30 to 50 children, with prototypes we produce ourselves. This is a plan, not traction. Investors: write to us for the one-pager.

NowClosed pilot, one child

Running at home on an off-the-shelf device with the sensor we need. Worn all day, raw signal, labels. About 15 episodes. The question: can we warn ahead, at home?

Step 2Closed beta, six families

Six families are ready, from a parent community of 135 families. Same protocol, more children.

Step 3Protocol and regulator

A scientific advisor turns pilot data into a protocol. Ethics, consent, data rights. FDA Pre-Sub before any claim.

Step 4Prototype production

Our own small batch of bands that give raw signal and that children keep on. Research devices are gone from the market.

Goal · 6 to 9 monthsStudy and paper

With clinical psychiatrists, on 30 to 50 children, by protocol. Then the package for a round.

Open questions we still have to answer: who pays, medical device or wellness, and what the alert threshold should be.

FAQ

Short answers.

What is Emozu?

A research project. We test whether a soft band on the wrist of an autistic child can warn a parent minutes before a meltdown, at home.

Is it a medical device? Can I buy one?

No. Emozu is at the research stage. It is not a medical device, it does not diagnose or treat anything, and it is not for sale.

How can a band know a meltdown is coming?

Published studies see the body change 1 to 10 minutes before an episode: heart rate, skin conductance (EDA) and motion. A model learns that pattern and sends a quiet message to the parent.

What does the band measure?

Heart rate and HRV, skin conductance (EDA), skin temperature and 3-axis motion. There is no screen on the child and no alarm in the room.

How accurate is it?

Published models reach AUROC 0.71 to 0.87, where 0.5 is a coin flip. The one study that reports alerts found about 2 of 3 were false. So we start silent: record, label, look back. Live alerts come only if the numbers hold.

Why a little friend on top?

Many children pull a medical sensor off. A friend they like is easier to keep on. The friend also holds the battery, so a child can swap friends and keep the band on.

How can my family join?

Right now the pilot is closed: one child at home. Next come six families from a parent community of 135. Write to us through the form. The form is for adults, and please do not send medical details there.

I am a clinician, researcher, engineer or investor.

Write to us. We are looking for clinical psychiatrists for the study, hardware and signal engineers for the prototype, and investors for our pre-seed round.

For parents

All articles

Write to us

One button. Tell us who you are and what you have.

We need people who can turn a pilot into science and a product. Investors, clinicians and psychiatrists, hardware engineers, research groups, families. Write a line, drop a link, or attach a file.

Yury, the founder, reads every message.

Prefer email? [email protected]

Emozu on LinkedIn