Engineering a sophisticated in-house risk management solution for a commercial lending platform doesn't necessarily need to involve millions of lines of low-level code, clunky desktop applications, fancy front-end development, or messy spreadsheets. That is, so long as the problem is approached objectively and the solutions are evaluated critically.

This talk will focus on the basics of lending risk mitigation (related to underwriting and portfolio management), a high-level overview of the architectural requirements, and the packages that were leveraged along the way; namely:

• pandas for data preparation
  • Organizing panels and cleaning time series data
• statsmodels and scikit-learn for regression and classification
  • Predicting accounts receivable discontinuation and bankruptcy probabilities
• R/rpy2 for integrating R's advanced forecasting capabilities
  • Forecast comparison using the forecast package and the bsts package
• Spyre for designing a lightweight and easily maintainable web application

We will share a demo of the web application to see how it all comes together, but to be clear, this IS NOT a product pitch or any kind or SaaS; it was built to be used internally and generalizes well for other use cases.