Vinitra Swamy - Building AI That Teaches: The Learning Science Behind Adaptive Tutors

​Large language models make it possible to build learning systems that do much more than answer questions. They can model what a learner knows, decide what they should encounter next, generate practice and explanations in real time, and adapt based on where they struggle.

In this talk, Dr. Vinitra Swamy will introduce Scholé, an multi-agent learning platform built around ideas from decades of learning science: retrieval, mastery learning, the zone of proximal development, knowledge tracing, active and constructive learning, feedback, and productive struggle. She’ll show how these ideas translate into the architecture of an adaptive learning system, from learner models and knowledge graphs to pedagogical agents that determine what happens next.
The talk asks a larger design question for AI education: What should we personalize, when should the system intervene, and how can we build AI that helps people learn rather than simply making it easier to get an answer?

Vinitra Swamy is the co-founder and CEO of Scholé AI, an UC Berkeley and EPFL lab spinoff developing AI-native personalized learning for the modern workforce. She is the first PhD graduate of the ML for education lab at EPFL, where her work on “A Human-Centric Approach to Explainable AI for Personalized Education,” was awarded the Patrick Denantes Prize for EPFL’s best computer science thesis. Before founding Scholé AI, she spent two years at Microsoft AI as a lead engineer on ONNX, collaborating with more than 30 technology companies to help establish an industry standard for AI interoperability. She also serves as teaching faculty for Harvard Data Science Initiative’s Agentic AI Intensives, and previously taught machine learning at UC Berkeley and UW.