Giving Foundation Models a Notion of Now
Exploring how encoding transaction timestamps as time deltas rather than absolute dates boosts AI model performance and generalization.
Research at Nubank drives innovation by deeply understanding the needs and challenges of our customers. Our team constantly explores new technologies and methodologies to improve our products and services, ensuring that we deliver exceptional experiences. Through data-driven insights and user-centered design, we are able to stay ahead of trends, offering smarter solutions that truly make a difference.
Exploring how encoding transaction timestamps as time deltas rather than absolute dates boosts AI model performance and generalization.
We explored how the Muon optimizer, an innovative alternative to AdamW, helps us build more efficient foundation models, with faster convergence and reduced costs.
Nubank leaders share how AI is shaping products, teams, and decisions. DS and MLEs play a key role in building the purple future. Read more in this blog post from Purple MinDS.
Instead of seeing AI as a tool just for automation, we view it as a catalyst — unlocking smarter decisions, freeing up time for more meaningful human interactions, and helping us scale without losing the essence of who we are.
Nubank optimizes data representation for its AI models by treating information selection and its representation as a hyperparameter search. This approach reduces the effort required to incorporate new data and significantly improves model performance.
We’ve advanced in adopting Predictive Foundation Models on our AI Platform, enabling large-scale sequence model decisions. This is a key step toward Nubank’s AI-First vision.
Careers
We bring together great minds from diverse backgrounds who enable discussion and debate and enhance problem-solving.
Learn more about our careers