Current
Working on ML/AI for battery energy storage operations at FlexGen.
Previous
Plastic Pollution Spectral Modeling
Collaborated with researchers at the University of Toronto to address the challenge of characterizing paint microplastics in the environment. Paint is modeled to be the largest source of microplastic pollution globally. We created a comprehensive Fourier transform infrared spectroscopy (FTIR) library named the Paint Library of Plastic Products (PLOPP), including 263 spectra from 90 different paints. Machine learning techniques were applied to classify spectral data as paint vs. non-paint microplastics.
EnergyScore
Solstice operates EnergyScore, a machine learning algorithm that delivers a more effective and more equitable qualification metric calibrated for renewable energy financing. Through support from the Tides Foundation and Google, I investigated the degree of bias existing in the algorithm, conducting a large literature review to identify leading quantitative measures of bias and fairness in ML.
Paper: Fairness in Focus: Quantitative Insights into Bias within Machine Learning Risk Evaluations and Established Credit Models
Repo →
Google.org subsequently provided a grant to Solstice to fully develop the ML product using these bias metrics. We collected an updated dataset of over 30 million account-level records.
Churn Model
At Solstice, developed a churn prediction model for community solar customers using logistic regression, random forest, and gradient boosting. AUC > 0.95. Deployed in production to identify customers at risk of leaving.
- Can AI help community solar projects hold onto their customers? — Canary Media
- Community solar developers look to AI to manage subscribers — Energy News Network
- Inside community solar's foray into machine learning — Latitude Media
SETO
Multi-tiered analysis of community solar prioritization and barriers to LMI access, funded by the US Department of Energy's Solar Energy Technologies Office.