Rafael Celente

Company post describing nuFormer, a transformer-based foundation model for financial transaction sequences, and how the research (including work by Rafael Celente) is applied to production use cases like credit and recommendation.

Academic paper introducing an 8-bit (blockwise) quantized variant of the Muon optimizer, demonstrating memory savings and empirical/theoretical analysis of quantization robustness for large-model pretraining.

Nubank engineering blog post describing experiments using the Muon optimizer to improve data and compute efficiency when pretraining transaction-based foundation models; shows faster convergence vs AdamW.

Personal blog posts on Medium about aerospace engineering, data and machine learning, entrepreneurship, and space/rockets.