Research + Product
Technical deep-dives, biosensor science, and product updates from the people building cycle intelligence from the wrist up.
Inferring hormonal phase from a wrist biosensor requires bridging two hard problems: the signal is indirect, and the biology is noisy. This post walks through the model architecture we use and the calibration choices that determine prediction confidence.
HRV varies across the cycle in ways that are consistent enough to inform phase inference but subtle enough to require careful baseline modeling. Here is how we handle that tradeoff in practice.
We opened our early access program to fertility clinics and women's health platforms last quarter. This post explains the onboarding process, what data flows you get, and what feedback shapes the product roadmap.
Thermal signals at the wrist are not driven by a single hormone. Cortisol modulates estrogen-driven temperature changes in ways that matter for inference accuracy. Understanding the interaction is essential for building robust models.
Not all wearable datasets are equally useful for cycle phase modeling. Dataset quality depends on sampling density, label methodology, and demographic range. This post covers what we look for and the gaps that still exist in public datasets.
Fertility clinics schedule interventions around cycle timing. Continuous passive phase inference from a wearable can reduce reliance on lab-draw scheduling and give clinicians a richer picture of where a patient is in her cycle before each visit.
Wrist temperature provides a continuous ovulation signal that core-body basal methods cannot. We walk through what the signal looks like around the LH surge, how our model handles it, and where the accuracy limits are.
A phase prediction without uncertainty bounds is harder to act on than one with them. This post explains how we generate confidence intervals from our LSTM outputs and how clinical users can interpret them in practice.
Wearable biosensor streams are rarely regular. Gaps from device removal, motion artifacts, and variable sampling rates all affect sequence model performance. This post covers the approaches we use to maintain prediction quality under real-world conditions.
The luteal phase is often treated as a fixed 14-day window, but in practice it varies significantly across individuals and cycles. Understanding that variability matters for both prediction accuracy and how we communicate phase boundaries to users.
Wrist wearables measure skin temperature, heart rate, HRV, and movement. They do not measure hormones directly. The gap between what is measured and what we infer is where all the modeling work lives.
Circadian rhythms and menstrual cycles interact in ways that affect both the signals we collect and the inferences we draw. This introductory post explains the overlap and why it matters for building accurate cycle phase models from wrist data.
