Paras Koundal

Hi there 👋🏾 I am a physicist and researcher exploring the intersection of cosmic-ray physics and machine learning. I currently work at the University of Delaware, collaborating with the IceCube Neutrino Observatory in Antarctica — a one-of-a-kind experiment buried deep in the polar ice, designed to detect some of the most energetic particles in the universe. My work revolves around using modern ML techniques — especially Graph Neural Networks and deep learning architectures — to study the cosmic-ray energy and mass spectrum. Beyond model building, I enjoy thinking about how physical symmetries, domain knowledge, and data-driven methods can coexist. In practice, this means creating physics-inspired ML pipelines that are efficient, interpretable, and, ideally, a little elegant.
Before moving to the U.S., I was a doctoral researcher at the Institute for Astroparticle Physics, Karlsruhe Institute of Technology, Germany, where I began exploring graph-based representations of air-shower data and simulation-driven analyses. I earned my BS–MS in Physics from the Indian Institute of Science Education and Research (IISER), Mohali, where my curiosity for the universe first collided with the power of computation. I am a strong believer in the power of open-source and reproducible research.
I was born in the small town of Nagrota Bagwan, in the Kangra district of Himachal Pradesh, India — a peaceful valley cradled by the Dhauladhar Range. Growing up surrounded by mountains probably explains my fascination with vast, difficult-to-reach places — from remote observatories to abstract spaces in data.

Research Work

I spend most of my time analyzing data from IceCube/IceTop, building machine learning pipelines that reduce our dependence on large simulations, and searching for new ways to connect physical insight with data-driven inference. I’m particularly interested in:

Graph Neural Networks for particle-interaction and event-reconstruction problems,
Designing interpretable ML architectures for physics experiments,
High-performance computing for large-scale ML in experimental science,
And, increasingly, in how AI can accelerate discovery without replacing human intuition.

Before This

Over the years, I’ve worked in Japan (J-PARC, KEK), at IISER Mohali on neutrino flux estimation, and briefly in AI industry settings developing recommendation systems. Each of these experiences — from soldering detectors in the lab to training neural nets on supercomputers — shaped the way I think about research: as a blend of curiosity, craft, and computation.

Beyond Physics

Outside work, I’m fascinated by good writing, mountain trails, and the art of building systems — whether in code, research, or life. I believe science is most meaningful when it stays human; that’s probably why I still enjoy telling stories about it.