Aevum · aging-trajectory engine · Singapore

We didn't fine-tune a model.
We built one.

modelled trajectory
45.8 yrs
pace 1.11× · recalculating
Psychedaily
Somacontinuous
Bioperiodic
Aevumfused trajectory
live · synthetic
← 90 days three cadences, one rate now
Zero external LLM in the loop

Aevum doesn't catalogue the genome. It models the living human — every day — and it was trained, not prompted.

Velya Pulse is how the engine sees. Velya Twin is how it speaks. Both exist to feed the same core.

The insight

Two people with identical bloodwork and genetics can age at different rates.

The variable is largely psychological — stress load, emotional regulation, recovery. It moves daily, it compounds, and almost nothing in longevity measures it.

Every longevity app reads the body, ignores the mind, then nags you to sleep and walk more. Reading both, continuously, per person, isn't a feature you bolt onto an app.

It needed an architecture nobody had built.
So we built it.

The data · three living streams

Catalogue-first AI sees a snapshot.
Aging is a rate.

Modelling a rate needs longitudinal, per-person, intervention-linked data — the one thing catalogue-first AI structurally lacks.

01

Psyche

daily · psychometric + behavioural

Stress archetype, emotional regulation, social contact, sleep architecture, allostatic load. The stream nobody else carries — and the one that moves fastest.

Social / relationalIsolation and loneliness signals, contact frequency and pattern.
Sleep architectureREM and deep ratios, night-to-night consistency — not duration alone.
Allostatic loadCortisol rhythm, HRV, inflammatory markers — the running cost of holding steady.
02

Soma

continuous · wearable

Recovery, strain, cardiovascular and movement signal — read off the watch a member already owns. No proprietary hardware, no new thing to wear.

03

Bio

periodic · clinical

Bloodwork and clinical biomarkers drawn through partner clinics, anchoring the daily signal to hard biology at a slower, deliberate cadence.

Architecture · v1

Inside the core.
Our algorithm, not a wrapper.

Everything speaks one vocabulary: the hallmarks of aging. Signal enters left, trajectory leaves right.

psyche ENCODER soma ENCODER bio ENCODER gated fusion aging state TEMPORAL RECURRENT counterfactual WHAT-IF A B
01

Stream encoders

Each stream learns its own representation before anything is mixed.

02

Gated fusion

Weights each stream per person, per day. Not an average — a learned gate.

03

Aging state

A recurrent module carrying trajectory forward. Rate, not snapshot.

04

Counterfactual

Models the effect of an intervention before a member commits to it.

No external LLM sits anywhere on this path.

Most health-AI companies place a foundation model at the centre and wrap it. There is no such node here. Aevum was trained on longitudinal structure, and the reasoning stays inside our own weights — which is also why it can be held to a trajectory rather than a plausible sentence.

The instruments

Two surfaces. One engine.

The daily sensor

Velya Pulse

The consumer layer — hardware-free, on the watch a member already owns. Aria builds the day around modelled energy, and every action taken deepens the profile underneath it.

feeds Aevum · psyche + soma, daily
The simulation surface

Velya Twin

Where the counterfactual engine becomes visible — members and clinicians see the modelled effect of an intervention before choosing it. Specified and prototyped, not sold.

reads Aevum · population + per-person
Pulse SIGNAL IN Aevum TRAJECTORY + WHAT-IF Clinic partner REAL INTERVENTION Measured outcome LABELLED PAIR INTERVENTION-RESPONSE RETURNS TO THE ENGINE
Velya

Longevity, redefined for the people building the future.

Aria is a wellness guide, not a clinician · answers reflect modelled, pre-launch work