semarx
The Human Digital Twin (HDT)
An AI companion that understands what you are waiting for
AI companions remember what you tell them. To help with a real decision, they need to know more: whose response matters to you, what came back, and what is still unresolved. We are building the architecture and first prototypes of a Human Digital Twin that does this.
What It Is For
People don’t make decisions in one place. One choice, such as leaving work early for a child’s school play, can matter to an employer, a family, friends, and a personal standard learned long ago.
​
The Human Digital Twin is designed to help a person see a decision coming, understand whose response matters and why, notice what is still open, and decide what to do next: follow up, let go, or look for a response somewhere else.
​
It supports the person’s own thinking. It does not steer them towards goals chosen by the system.
What Is Different
-
Episodes, not profiles. Most assistants build a profile of interests and habits. The Human Digital Twin records episodes instead: what you did, whose response you expected, what actually came back, and how you took it (settled, partly settled, or still on your mind).
-
Two views, kept apart. The people and places a person answers to (work, family, friends, communities, and standards learned early), and the person’s own view: what they expected, how much it meant, and what is still open.
-
A clear difference. Current Human Digital Twin research mostly models the body (health signals) or task performance (workload, attention). We model how a person’s actions connect to the responses that matter to them.
A clinical instrument for neonatal intensive care. Concept and proof of concept.
​
Running alongside care on a unit’s own treatment data, the Patient HDT gives the team three readings in real time, each with a calibrated confidence:
-
Predictive: where a child is heading.
-
Diagnostic: what the current picture most likely means.
-
Treatment-related: which recovery path has worked before in similar cases.
-
The team reads, the clinician decides.
​​
Why it matters and how we build it: in neonatal intensive care, small changes escalate quickly, and knowing how to read them often sits with the most experienced clinicians. We represent care as state → intervention → outcome, learned from the unit’s real practice, and build a readable reference per condition, not a black box.
-
Basis: Transforming AI Solutions in Healthcare — The Medical Information Tokens, IEEE Conference on Artificial Intelligence (CAI), 2023.
-
Current step: a retrospective, pseudonymised proof of concept for 2–3 critical conditions, such as neonatal sepsis.
-
Analysis only: no bedside device and no change to care.
-
We are looking for clinical partners for the proof of concept.
The Patient HDT
A New Layer of Human–System Intelligence
Principles and Where We Are
-
How it behaves: the person leads, and every interpretation is visible and correctable (“You read that as seeking approval. I was keeping a promise.”). It asks questions rather than giving diagnoses, and it never replaces the response someone is waiting for.
​​
-
Privacy by context: what is shared in one part of life stays there. It is never used for advertising, hiring decisions or monitoring. It is built on consent, collects no more than it needs, and can be deleted at any time.
​​
-
Where we are: we are developing the architecture and first prototypes. The core is the representation: describing episodes, the people and places a person answers to, and how settled each expectation is, in a form both the person and the system can read and correct. We then test whether it helps people decide, not just whether it predicts their next choice.
​​
-
Work with us: we are looking for research partners, early users, and clinical partners for the Patient HDT proof of concept. Contact: idt@semarx.com