Data scientist and founder. Credit and lending ML.
I have spent 12 years building ML models that decide who gets credit, how much, and at what price. Today I run Go Fig, an AI company that reached $300K ARR.
Models with money on the line: Capital One auto lending, Oportun consumer credit.
Bank transaction data as an alternative source for applicants with short credit histories.
Production ML I own end to end: forecasting, anomaly detection, NLP and LLM agents.
Selected work
Go Fig: ML functions behind an AI agent
Founder and CEO, 2024 to present
Problem
Teams need forecasts and early warnings from their own financial data, without waiting on an analyst.
Approach
Built machine learning functions that our AI agent runs against a company's financial data. They forecast results and alert on anomalous spending. Deployed to production with monitoring and tests.
Result
The company grew to $300K in annual recurring revenue.
Oportun: credit strategy rebuilt around ML models
Analytics Manager, Risk Strategy, 2020 to 2022
Problem
Customer behavior shifted through the pandemic, and the credit strategy no longer fit the ML models underneath it.
Approach
Rebuilt the strategy around the models. Wrote an automated Python program that applies predictive models to recent volume and risk trends, which moved reviews to a monthly cadence. Used bank transaction data as an alternative source for applicants with short credit histories.
Result
Loan volume grew 30% with risk and yields held flat.
Capital One: valuation and decisioning models
Senior Business Analyst, Application and Decision Science, 2018 to 2019
Problem
Valuation and decisioning models steered $25B of annual auto originations.
Approach
Owned model performance and the product experience. Project managed the rebuild of the valuations model for higher predictive accuracy.
Result
More than $100M in added annual profitability, from lower net losses and higher volume.
Square: seller lifetime value from support calls
Senior Data Scientist, Growth, 2022 to 2024
Problem
Which sellers will be worth the most, and does the quality of a support call change that?
Approach
Built an NLP and Random Forest classification model that quantifies the lifetime value a seller adds based on support call quality.
Result
A multi-variant marketing experiment returned more than 30x ROI, and a growth roadmap drove 4x growth in new value added over 12 months.
Classifies bank transactions, finds recurring streams, estimates annual revenue with a range, forecasts with seasonality, and flags anomalous spend. The README reports where it falls short as plainly as where it works.
