Post by Edson Casimiro

Last week I attended London PyTorch Meetup 26 at Revolut HQ, where we had four back-to-back talks on where production AI is actually heading. Foundation models in different domains are shaping the way AI is being adopted by organizations. 🏦 Anton Repushko from Revolut introduced PRAGMA, a foundation model for banking event sequences, trained at scale across multiple financial tasks 🌲 From Trees to TFMs - this was the case for Tabular Foundation Models replacing tree-based pipelines in production, presented by Nicolas Atienza from Neuralk ⚙️ Ivan Dolgov form JetBrains explained how the company is building Mellum2, a MoE model co-designed for H100/H200 GPUs 🤖 And for businesses considering local LLM deployment, Kir Zharov talked about Unified Knowledge Layer as the infrastructure enabling self-improving agents This isn’t just research anymore. It’s production. Follow Edson Casimiro for more insights on R&D and AI implementation. #PyTorch #MachineLearning #FoundationModels #MLEngineering #LondonTech