test particles orbiting a black hole

r₀initial orbit radius
Langular momentum — petals ↔ circular
view
spinrotate camera — 0 holds still
zoomdrag empty space to turn

Art, physics, and software.

Software

Vehicle range predictions & route planningMachine-learning route forecasting — see how much range a drive will cost you before you set out.

Plan a route and it forecasts a quantity that changes as you travel. Built first for electric vehicles — predicting how much battery a given route will actually cost — but the model isn't tied to batteries: give it any target you can measure along a path and it learns to predict that. Route planning and GPS analysis are built around that prediction.

General Relativity

sym_grSymbolic General Relativity — follow the steps from a metric ansatz to the field equations.

An interactive symbolic-GR workbench built on SymPy. Start from a metric ansatz and apply physical assumptions one step at a time — static, spherically symmetric, and so on — watching how each assumption reshapes the metric tensor and the differential equations it implies. It derives the Christoffel, Riemann, Ricci, and Einstein tensors exactly, extracts the Einstein field equations, and lets you test a candidate against known solutions (Schwarzschild, for instance, collapses the vacuum equations to 0 = 0). Solve those equations and you have a complete metric — which feeds straight into GR below.

sym_gr live app — Einstein Field Equation builder
sym_gr · live app

reading live usage…

GRGeodesics from any metric, by symplectic integration.

Give it a metric and it returns the spacetime trajectories. Geodesics are treated as Hamiltonian flows and integrated with a structure-preserving symplectic scheme (following two papers) across an extended phase space of 4-position and 4-momentum — high-order error control and stable accuracy over long timescales, even for non-separable Hamiltonians. Define any metric symbolically, let it solve the energy component of the initial momentum from conservation laws, and push many test particles through at once. The natural partner to sym_gr: solve for a metric there, integrate its geodesics here. Below: geodesic orbits around a Kerr black hole (M=1, a=0.5), 4th order.

Geodesic orbits around a Kerr metric, 4th order
Kerr metric · geodesic orbits

Tutorials

From scratchMachine learning built up from the math — no framework magic.
  • neural nets how nets (and graphs) carry the algebra of backprop, from scratch.
  • glm generalized linear models built up from the exponential family.
  • knn k-nearest-neighbors implemented from nothing but distances.
  • taylor vae a variational autoencoder, derived and coded from the ground up.

Free to read and use. If they help, you cansupport the work →.

Live

an AI-assisted tool I built to organize, tailor, and send my applications.

approval rateof the applications the AI drafts, the share I approve to send
hands-offshare of each application the AI preps end-to-end — found, tailored, drafted — before I approve and edit or send