Examples#
End-to-end models of real decisions, each built step by step from the
runnable source in examples/. Every example runs on the free tier, and
every figure on these pages is computed by the model it illustrates.
Energy and power#
A meshed grid under uncertain wind and demand: PTDF and LODF from the gradient of the DC power flow, N-1 screening, cascading overload trips and line uprates optimised by gradient.
A radial feeder where faults cascade along lines; which lines to reinforce and where to put the substation are optimised in one gradient pass.
Blade element momentum for the rotor, a directional wind climate, wake losses across the farm, the AEP distribution with P50 and P90, and a layout optimised by gradient.
Transport and structures#
Lift and stall of a light aircraft in turbulent air: gust load factors, stall probability, the lift curve calibrated to wind-tunnel data and the sensitivity of the risk.
A coupled wind process with a circular, bimodal direction drives crosswind exceedance and wind-shift risk for two runways.
Rainflow-counted fatigue of a welded detail under simulated traffic, calibrated to strain gauges, with the reliability over a 100-year life and a flange designed by gradient.
Environment and fluids#
A Gaussian plume from a leak anywhere along the route, wind as random inputs, exposure and its tail, a calibrated dose response and abatement optimised by gradient.
Theis drawdown of a pumping well field with uncertain aquifer properties calibrated to observation wells, the risk of drying out a fen and pumping rates optimised under that risk.
Life sciences and insurance#
Between-patient drug kinetics learned from sparse, censored monitoring samples with an exact compartment model, and the loading and maintenance dose designed by gradient.
The exact aggregate loss of a portfolio by Panjer recursion, layer pricing, capital gradients, an optimal retention and a Monte Carlo and copula cross-check.
Learned surrogates#
A network surrogate of a battery thermal model inside the risk model: pretraining, extrapolation, fine tuning to field data, attribution of the risk and export.
A whole model’s predictive distribution amortized into a fast surrogate, validated against the model’s own Monte Carlo error, with guarded fallback to the full model.