Developer builds full-stack credit risk system in Python using 1.35M real loans

A software developer has built a comprehensive credit risk framework in Python using 1.35 million real Lending Club loans, going well beyond the default prediction models covered in most tutorials. The project covers three connected components: an IFRS 9 Expected Credit Loss engine, a Weight-of-Evidence scorecard with independent model validation, and a portfolio monitoring pack. Key techniques include out-of-time validation using older loan vintages for training and newer ones for testing, and measured Loss Given Default rather than assumed values. Applied to a live book with $9.5 billion in exposure at default, the system produced an ECL of $1.25 billion and an overall coverage ratio of 13.1%. The entire stack relies on standard Python libraries — pandas, scikit-learn, and matplotlib — making it deliberately reproducible for practitioners.