Sudip Roy
Co-founder & CTO
Adaption
Sudip Roy is a Co-founder and CTO of Adaption, where he leads engineering efforts to build intelligence that continuously evolves. Sudip was previously the Senior Director of Engineering at Cohere, where he led Inference, Finetuning, and Platform for the company’s foundation models. During his tenure, he scaled the engineering organization tenfold while helping drive revenue growth from under $1M to $100M through industry-leading advances in system efficiency and reliability. A specialist in both systems for AI and AI for systems, Sudip spearheaded strategic technical collaborations with major partners including Nvidia, Google, Oracle, Amazon, Microsoft, and Groq to optimize AI deployment at scale. Prior to Cohere, he was a researcher at Google DeepMind, where he contributed to the core ML infrastructure—such as Pathways and TensorFlow Extended—that powers training and inference thousands of models in production including Gemini. He holds multiple patents in machine learning systems and is a recognized leader in scaling foundation model infrastructure.
Schedule
15 mins
Meridian Hall — Main Stage
Beyond Brute Force: Building AI that Adapts
The next leap in intelligent systems may not come from bigger models, but from their ability to adapt. This session explores how AI can continue learning after deployment, evolving alongside changing conditions and closing the gap between what a model was trained on and what businesses face in the real world. Join us as we unpack what often breaks between AI pilot and production, why continuous learning can be more effective than simply scaling models, and how organizations can build systems that improve from the data and traffic they already handle. The goal: AI that works for people, rather than asking people to work around it.