ml · materials
BREED 2.0 — ML for solid-state electrolytes '25
DFT screening of electrolytes is brutally slow. The hard part: predicting ionic conductivity from composition alone, with almost no labelled data. Built a gradient-boosted model + genetic algorithm to search the candidate space.
MAE ≈ 0.94 (log σ) on 829 samples
beats the published OBELiX RF baseline by 36%
beats the published OBELiX RF baseline by 36%
controls
EKF to estimate battery state-of-charge '24
SoC isn't directly measurable and voltage is nonlinear + noisy. The hard part: ran an Extended Kalman Filter over an equivalent-circuit model on an 8-bit Arduino.
nonlinear estimation, on-device
