About

Simon Roth

PhD, Graduate School of Decision Sciences, Konstanz (2022) · Independent Researcher

I build ML tools where structural errors can't hide. The ML grammar is the result: a typed workflow for Python, R, and Rust where the most common validation mistakes are unrepresentable by construction.

8+ years of ML pipelines, data products, and frontends. Before that: quantitative social research and too many regressions. The PhD was in Decision Science — which meant spending several years studying how and why analytical workflows produce conclusions that don't hold. Mine included.

One-person lab. Building with AI agents as research partners — not as novelty, but because the combination of human constraint-writing and automated adversarial stress testing at scale produces more rigorous coverage than either alone.

Code ships under MIT license. Papers go on arXiv. Kill criteria are recorded before every experiment runs. When the data says an idea is wrong, the negative result gets published.

Outside the lab: gardening, cooking, philosophy, programming, podcasts, people. Multidisciplinary by instinct — most interesting things happen at the edges. Epagogy style: observe, hypothesize, test, revise. Repeat.

Epagogy — επαγωγή — Aristotle’s term for induction: reasoning from particular observations to general constraints. Prior Analytics, II.23. The domain was free. The philosophy was a coincidence that turned out not to be.