At Notzero Technologies, I built Recoup, a payment-recovery system that chooses when to retry failed subscription payments and when to request a new card. In simulation, it recovered 31% more revenue per failed payment than the fixed retry rule.
I shipped Recoup as an operations CLI and a FastAPI and React operator console. I modeled payment likelihood at each retry time with a mixture-cure survival model in Python and SciPy, including an approach pooled across merchants.
My projects include Can I borrow.ai, a loan-default risk engine with calibrated, versioned models served through a tested Flask API, and PyTorch LSTM and GRU models for forecasting. I also built a creator brand-deal platform MVP and a tested Python package for retail sales predictions.
