Autonomous improvement
Systems that generate hypotheses, test them against objective feedback, and find improvements humans have not found.

I'm Pragun Seth, a founder and engineer/researcher at Berkeley building self-improving systems.
My interests center around solving problems where AI can accelerate discovery and engineering for public benefit.
How do we build intelligent systems that learn, experiment, and discover better solutions on their own?
Systems that generate hypotheses, test them against objective feedback, and find improvements humans have not found.
Memory, continual learning, evaluations, post-training, and agents that get better through deployment.
Robotics, manufacturing, energy, and autonomous engineering in the physical world.
Open technical frontiers where strong intuition and agency matters.
Currently, I'm focused on developing deep technical skill across agentic discovery and verification.
My goal is to answer this question: what does the world need next?