The intelligence layer for mental health

Before Carl Wunderlich, fever was judged by a hand on the forehead. In Leipzig he took readings from tens of thousands of patients, drew what normal looked like as a curve, and showed that the shape of a fever over days told you more than its height at any one moment. Temperature stopped being a feeling and became a line.

Every protocol medicine has for fever descends from that line. Blood pressure followed, with the cuff. Glucose followed, from quarterly lab draws to a sensor on the arm that reads every few minutes. Cardiac rhythm followed, from an annual trace to a wrist that listens all night. The rule has held for a century and a half: continuous measurement comes first, and effective treatment follows it. No part of medicine has ever gone the other way round.

Mental health is still where fever was before Leipzig. The instruments that decide whether someone is referred, what they are offered, and whether anyone notices a change are short questionnaires, answered once, from memory, about the past two weeks. They measure what a person can recall of how they felt. They cannot measure how a person is.

Set that beside the rest of the body. The heart gets a reading every second; the mind gets a form every few months, if it gets one at all. An entire domain of medicine runs on someone rating their distress out of ten, in a room, on a Tuesday. And it is the domain driving much of what goes wrong in the rest of the body.

Asking was never a mistake. There is no blood test for how someone feels, and for most of medicine's history a conversation was the only instrument that could reach it. Clinicians built careful, validated ways of asking, and in a consultation, with a trained person interpreting the answer, they work.

They stop working the moment the person leaves the room, because a question cannot be asked continuously. Meanwhile, the things that shape mental health leave traces long before anyone puts them into words. For the first time many of those traces are recorded by default, by the devices people already use, without anyone having to ask. The raw material for a Wunderlich curve of the mind now exists at population scale. What does not exist is the system that reads it.

Mental healthcare today begins after deterioration has already happened. A person feels worse, recognises it, decides to ask for help, is assessed, and only then does treatment begin. By the time the system sees the problem, the problem has often had weeks, months or years to deepen.

More treatment matters, but more treatment alone cannot solve this. The later we intervene, the more severe the problem tends to be and the more skilled, intensive and resource-heavy the care required to change it. Specialist capacity will always be finite, and it takes years to build. A system that waits until large numbers of people need specialist care will always be running behind demand.

The answer is not simply to make the end of the pathway bigger. It is to move the beginning of care earlier.

There is a window before deterioration becomes severe where a person may still be able to change course through sleep, movement, routine, social connection, guided self-management, digital support, lower-intensity care or other interventions that do not require specialist treatment. But that window only matters if the change can be seen while it is still open.

You cannot intervene early in something you cannot see.

Measurement is what makes prevention operational.

The first attempts to read the mind with software each chose a single channel: the words in a transcript, the tone of a voice, the rhythm of typing, a conversation with a chatbot. One channel gives one view of a person, and one view is a high-noise guess.

Mental health is not one signal. It is a whole life, and reading any part of it in isolation misses what matters.

This is how we think about AI. A language model reading a transcript is a single sensor. What we are building works above the sensors, at the level of the whole person.

A person connects what they choose to share, and every source they add sharpens the picture. What comes back is a measurement that belongs to them and moves as they do.

But the point is not measurement for its own sake. It is to make the window before illness becomes severe visible and actionable.

Caire explains what it sees, suggests what to try and follows what happens next. Over time it learns not only what changes a person's mental state, but what changes it back.

This helps people understand what shapes how they feel and act while the problem is still small enough to change. Measurement comes first because everything else is built on top of it.

The inputs are all but limitless. Last night's sleep, this morning's argument, the walk that did not happen, the message that did, the month of money worry underneath all of it. They act on each other constantly, and the same input can push a person in opposite directions depending on everything else that is true that day.

Our work is to make sense of all of it, for each person, and to know what can turn things around while there is still time to act.

The whole of this is one system. We call it the mental model.

Our founders met at Imperial College London while one was in medical school and the other was starting his PhD in computational acceleration in high energy physics, working with CERN.

Our founding team has spent their careers at the frontier of clinical medicine and machine intelligence: treating patients across psychiatry and acute medicine, running psychiatric research, building the first sub-microsecond inference engine for the Large Hadron Collider upgrade, and engineering the low-latency systems behind high-frequency trading, across Imperial College London, the NHS, CERN and Jump Trading. No other team sits deeper across both medicine and machine.

We came together around a shared conviction: that mental health will never be treated well until it is measured, and that the measurement has to be built by people who have sat on both sides of it. Our clinician founder left medicine to build it.

Mental healthcare today is reactive. It begins when someone finally asks for help, which is usually after the collapse.

The answer cannot simply be more care at the end of that process. Specialist treatment will always be finite, and the later we intervene, the more of it people need. A system that waits for deterioration to become severe and then tries to build enough capacity to treat everyone downstream will always be chasing the problem after it has grown.

We see a different model.

A world where the drift is caught the moment it begins, by the person themselves, long before it becomes a crisis. Where the first intervention happens while the person can still change course. Where self-directed care, technology and lower-intensity support deal with what they can, and specialist treatment is concentrated on the people who genuinely need it.

Where care is offered because a signal moved rather than because someone finally asked. Where treatment is chosen on evidence about this person, and its effect is visible immediately. Where the system learns what works for them. Where the baseline stops falling.

Better mental health for everyone. That is the whole mission, and it starts with a number.