Four careers, one problem

People occasionally ask why my CV reads like four careers at once: general relativity and early-universe cosmology, then black hole magnetospheres, then physics-informed machine learning for cardiology, then neutrinos, and now quantum-enhanced CFD and markets alongside all of it. It is a fair question, and the objects really are seemingly unrelated: anisotropic shear in a Bianchi-I universe; a lepton accelerator sitting outside a Kerr horizon; Sagittarius A* in general-relativistic magnetohydrodynamics (GRMHD); a catheter grid on a fibrillating atrium; a liquid argon time projection chamber (TPC); a futures curve; and, most recently, blood moving through a coronary artery. Roughly seven fields, and not one of them shares a journal with another. I take that as the good news: it means the thing they do share had to be something deeper than subject matter.

The honest answer is that I have only ever worked on one problem. You have noisy, incomplete data produced by a process you understand partially. You want a statement about the world that is not merely precise but calibrated - one whose error bar means what it claims to mean. Everything else is domain vocabulary. A small uncertainty is easy to manufacture and worth nothing by itself; what matters is whether the number is honest about how wrong it might be, because somebody is going to act on it. A cardiologist decides whether to put in a stent. A neutrino collaboration decides whether it has seen charge-parity violation in the lepton sector. Neither decision cares how impressive your resolution looked in isolation.

As for black holes, frame dragging near a Kerr horizon; for PINNs, conduction velocity across an atrial electrode grid; for neutrinos, a flux systematic propagating from a near detector to a far one; for markets, a regime shift in a futures curve. The objects differ; the inference structure does not. The shape that keeps recurring is a governing equation you trust, an observation operator you do not, and a decision waiting at the end of it. What varies is only which part you are least sure of.

The vocabulary gives the game away. In CT-FFR, the accuracy of the pipeline is not limited by how fast the solver runs; it is limited by the outlet boundary conditions, where an entire downstream vasculature you cannot observe gets compressed into a handful of lumped parameters. Make a thousand-fold faster solver and you converge on the wrong number sooner. In markets, impact, inventory, liquidity and latency are the boundary conditions, and a strategy specified without them has not been specified. Same phrase, and not by coincidence: in both cases it names the place where a clean interior problem meets a world you only partly observe.

Once you see this, moving between fields stops feeling like starting over and starts feeling like carrying the same toolkit into a new room. The furniture is unfamiliar, but the tools are very familiar.

So that is what this blog is for: the work, from inside it. What I am reading; what broke this week; and, now and then, the thing that finally worked. Expect few posts, and real ones.

It will be written entirely in English, as the rest of this website is. I am aiming for a global audience, and most of my network is based in Europe or the USA, so English is the one language that reaches all of it.