TimeTrace Labs builds the measurement infrastructure for healthcare: a maths-first layer for physiological signal extraction, paired with a frontier-scale world model trained on time-series.
We are a team of physicists, mathematicians, quantitative researchers, ML engineers, and neuroscientists from Oxford research roots, building calm, interpretable systems that track physiological drift across time and modalities.
TimeTrace Labs combines a maths-first measurement layer built for noisy, irregular biological signals with a world-model layer trained on physiological time-series.
The measurement layer provides interpretable, audit-grade descriptors at the individual level. The world model extends those descriptors into richer representations where the data supports it.
Together, they form a calm, dependable measurement infrastructure for healthcare that can track drift across diseases, modalities, and timescales.
Both layers operate on the same physiological substrate. The measurement layer constrains; the world model amplifies. Outputs are fused — interpretable descriptors paired with learned representations, calibrated together.
Built for noise, irregular sampling, and individual-level inference at low N. Every output is a calibrated descriptor — full provenance, audit-grade, interpretable by construction.
Representation learning trained on physiological time-series at scale — across modalities, populations, and timeframes. Amplifies signal where the data supports it; defers to measurement where it doesn't.
Raw multimodal time-series enter on the left. The measurement layer constrains noise; the world model lifts representation. Calibrated descriptors leave on the right — each one traceable to its origin.
Trials, studies, consumer devices and health systems — past or present, single or multimodal, disease-agnostic. The platform travels with the data.
Adaptive filters that respect physiological priors and survive irregular sampling.
Individual-level estimates at low data, calibrated to each subject's own baseline.
Slow change in physiology, surfaced before it becomes a clinical event.
EEG, ECG, actigraphy, glucose, voice — combined under a single representation.
World-model latents you can probe along clinically meaningful axes.
Every descriptor carries a confidence interval the world model can't override.
Every output traces to a primitive. No black boxes between physiology and decision.
Representations trained on one population that hold up on another, with proof.
Every call returns a versioned, signed record — built for clinical and regulatory review.
Spun out of a decade of research at the University of Oxford. We bring the discipline of fields that have learned to measure under exactly these conditions — particle physics, quantitative finance — and apply it to the conditions of biology.
Trials, studies, consumer devices, health systems — past or present, single or multimodal, disease-agnostic. We start from what you already generate.