This session explores the latest advancements in tabular AI, spotlighting cutting-edge solutions like TabPFN and CARTE. These models use transformers, synthetic pre-training, and novel data representations to rival—and potentially surpass—traditional gradient boosting and earlier deep learning methods. After a review of past approaches, such as TabNet and TabTransformer, Luca will explain why tabular data has remained a difficult domain for deep learning. Attendees will gain insight into the distinct architectures of TabPFN and CARTE, their benchmarking performance across various datasets, and their strengths with and without hyperparameter tuning. The session concludes by addressing real-world deployment challenges in enterprise environments, including model explainability, interpretability, and the promise of accelerated experimentation. Ideal for data scientists, ML engineers, and analysts eager to harness next-gen AI for structured data.