microsoft-research
(7)
Pavan Kalyan Tankala and the trouble with done
From expressive story voices at IIT Bombay to creative AI evaluations at TensorTest, Pavan Kalyan Tankala keeps returning to a demanding question: how do you know the work is any good?
Jiang Daxin and the wind that reached his door
After 16 years building AI products at Microsoft, a few conversations with ChatGPT sent Jiang Daxin into a new job: founder. At StepFun, he is trying to turn a researcher’s long view into models that people can actually use.
Greg Yang and the Small Model That Could Teach a Giant
Before co-founding xAI, the mathematician built a framework that lets researchers test training choices on small neural networks and carry them to far larger ones. His route there ran through a Harvard detour as a DJ, a prizewinning thesis, and a stubborn belief that deep learning should have a theory.
Pavan Kalyan Left the Lab to Make Pharma Paperwork Move at Machine Speed
An IIT Bombay researcher learned to make machines tell stories. Now he and two friends are teaching them the less glamorous, more consequential craft of regulatory documentation.
Abhay Agarwal
Abhay Agarwal is the founder and CEO of Poly, the intelligent cloud file browser built to replace Finder and File Explorer for the AI era. A former Microsoft Research AI fellow and Stanford d.school lecturer, he came up through engineering at UC Berkeley and design at Stanford, then founded the human-centered AI consultancy Polytopal before pivoting it into a Y Combinator-backed startup. In November 2025, Poly emerged from stealth with $8M in seed funding led by Felicis, powered by its own Polyembed-v1 embedding model. Agarwal describes the product as 'an LLM with infinite context from your life,' and frames his life's work around a theory he calls Post-Cognitive Design.
Andrew Mao
Andrew Mao is the co-founder and CEO of Parsnip, an AI-powered learning app often called 'Duolingo for cooking' that helps novices build real confidence in the kitchen. A Harvard computer-science PhD and former computational social scientist, he ran the longest iterated prisoner's dilemma experiment ever conducted, built ML systems at neural-interface startup CTRL-labs (acquired by Facebook), and now applies the science of how people learn to the very human problem of figuring out what's for dinner.
Ashish Kapoor
Ashish Kapoor is the co-founder and CEO of General Robotics (formerly Scaled Foundations), a Redmond, Washington startup building a unified intelligence layer for robots of every shape. After 17 years at Microsoft, where he led the Autonomous Systems and Robotics Research group, created the open-source AirSim simulator, and co-authored the widely read 'ChatGPT for Robotics' paper, Kapoor set out in 2023 to make robot intelligence portable across hardware. His thesis: general intelligence emerges from the rich composition of many robot skills, not from a single ever-larger model. He holds a PhD from the MIT Media Lab, flies and has built his own airplane, and treats simulation as the laboratory where machines learn to be safe before they touch the real world.