Data Scientist
CurrentExplainability and Fairness in Credit Underwriting-------------------* I participated in designing a product to assess and mitigate discrimination in lending. I researched fair lending laws, the current art in their enforcement, and the key areas of ambiguity in fair lending regulation. * I developed:(a) interpretable measurements to proactively monitor for discriminatory practices. (b) attribution of disparities to driving factors in a model.(c) automated suggestions of what an applicant can do to be approved in the future.* I lead our data science team's participation in a Stanford+FinRegLab research study: https://finreglab.org/ai-machine-learning/explainability-and-fairness-of-machine-learning-in-credit-underwriting/Sales Optimization-------------------* I helped our client maximize the profitability of their sales team's time. We did this by automatically deciding which sales people each inbound lead should be allowed to book an appointment with. * We used machine learning to maximize the expected revenue gain from our booking decisions whilst using control theory to ensure none of the salespeople have too few or too many appointments. We validated the logic with backtesting for profitability, appointment calendar stability, and prediction error. We also designed production a/b testing to monitor actual revenue gain and iteratively make improvements to the system.* We used python, scikit-learn pipelines, pandas, numpy, scipy, and our in-house Probabilistic Rules Engine. I built a custom controller in numpy to stabilize the supply of each sales person's time in response to demand in a manner similar to dynamic pricing.
