Master's student in Atmosphere/Energy at Stanford, in the department of Civil and Environmental Engineering. Before that, Ecole Polytechnique.

I work on optimisation, power systems and energy storage, and on the scientific computing and machine learning that sit underneath them: methods applied to real data, where a formulation that is sound on paper can still fail quietly on one input or on one machine.

single-diode fits the five parameters of a photovoltaic module's equivalent circuit from its published datasheet, then solves the circuit. The difficulty is numerical rather than conceptual: the textbook closed form overflows double precision on real modules, and whether the parameter fit converges depends on where it starts.

campaign runs a parameter sweep described in one YAML file, locally or as SLURM job arrays, and checks every run against invariants the physics says cannot fail. What it guards against is output that looks fine: a run killed mid-write that still counts as finished, or a result reused after its model file has changed.

cyclelife predicts how long a lithium-ion cell will last from its first 100 cycles, on the public MIT-Stanford dataset, splits strictly by cell, and replicates the paper's linear model to within five cycles. The first run was off by a factor of three on one test set: two physically impossible capacity readings, found by comparing each set with the paper separately.

Reach me at valmann@stanford.edu, or on LinkedIn.