TiRex-2 - a multivariate time series foundation model

In this episode, I sit down with Levente Zolyomi, a leading PhD researcher at NXAI, to unpack the next evolution in time series AI: TiRex-2. We explore how TiRex-2 builds on its predecessor by handling complex multivariate data and streaming scenarios, opening new frontiers for industrial forecasting. Levente shares the story behind TiRex-2’s architecture, its breakthrough capabilities, and what sets it apart from transformer-based models like Kronos. I ask the questions you’re thinking—about zero-shot forecasting, synthetic data, and real-world benchmarks—so you can understand what matters most for your business. If you’re navigating industrial AI or just curious about the future of time series models, this conversation is packed with the insights you need.