Independent validation of models, simulations and algorithms

An independent review that tells your team whether a model, simulation, AI pipeline or quantitative algorithm can be trusted for the decision it is meant to support, before it drives design, investment or deployment.

Why it matters Code can solve its equations correctly and still describe the wrong system, and a backtest can pass while hiding the assumption that breaks it in production. As simulation and AI tools make it faster to build models, checking their assumptions has become the bottleneck. I review physics-based models, multiphysics and device simulations, machine learning pipelines, and optimization or trading algorithms.

Method

1. Assumption audit that traces each physical or statistical hypothesis from the equations into the code.
2. Independent analytical or reduced-order cross-checks of key results.
3. Sensitivity and stability analysis, including limiting cases and edge regimes.
4. Comparison with measured, experimental or literature data where available.
5. Clear statement of the domain where the model is valid and where it stops being trustworthy.

Deliverables Written report with a verdict, ranked issues, annotated corrections or code fixes where needed, and recommended tests or measurements to close the remaining gaps. I can work directly in your codebase (Python, MATLAB, JavaScript) under NDA.

Turnaround and capacity Typically 1 to 4 weeks at a fixed scope, delivered remotely in English or French.

Background PhD physicist from CEA-Leti, where I built and validated compact models of FDSOI transistors down to cryogenic temperatures against Poisson-Schrödinger simulation and measured data (published in Solid-State Electronics). Recent work includes mathematical validation of nonlinear dynamical models and a trading system with historical backtesting and forward testing.

Best fit for R&D teams relying on simulation or AI for design decisions, startups whose core claim rests on a model, and investors who need an independent technical check.