Two questions. Deeply studied.
SandOaks is a small private research group. We apply machine learning to two domains we find genuinely worth a decade of attention — human health & diet, and financial time series.
What we work on.
Models that read the body.
We build models of human physiology — cardiovascular signal, sleep architecture, metabolic response — to study how diet and training change the body over months and years, not minutes.
Models that read the tape.
We build sequence models for financial time series — regime detection, volatility structure, multi-horizon forecasting — at scales where statistical edge is small and discipline matters more than cleverness.
Long time horizons. Small team.
Start with the signal, not the model.
Every project begins with months of just reading data — what is noisy, what is missing, what the textbook story leaves out.
Train models that hold up out of sample.
We optimise for calibration, robustness, and clean failure modes. A model that knows when it doesn't know is more useful than one that's right on average.
Put the model in front of a real decision.
We aren't here to write papers. Every model we train is wired up to a decision someone we trust is making this week.
A private lab, by design.
SandOaks is privately funded and intentionally small. We don't take outside capital, don't run a consulting practice, and don't publish in academic venues. The freedom to choose our problems — and the patience to study them slowly — is the point.
Health and markets are very different surfaces, but the underlying questions rhyme: how do you build a model of a system that is noisy, non-stationary, and partially observed — and trust it enough to act? That's the work.