Kimaya Mehrotra

Kimaya Mehrotra

Biography

Kimaya is a rising junior in Chemical Engineering at the University of Illinois, Urbana-
Champaign. Through her research, she aims to leverage the advanced logical and analytical
capabilities of computation to understand multiscale chemical systems. This summer, Kimaya is
working with the Laboratory for Atomistic and Molecular Mechanics (LAMM) at MIT, designing
an AI-driven closed-loop scientific discovery system for materials science. At UIUC, Kimaya
has been conducting research with the Peters Lab for the past year. She has contributed to the
development of the group’s Master Equation microkinetic modeling technique by designing kinetic
Monte Carlo simulations to capture the bistability phenomenon in chemical systems. Kimaya
values responsibility in engineering and understanding the ethical implications of emerging
technologies, and hence enjoys learning about the history and philosophy of science. She is also
actively committed to STEM teaching and outreach initiatives, helping young students experience
the inherent wonders of science.

Sparks2D: Multiagent AI for Autonomous Discovery in 2D Materials
Kimaya Mehrotra1, Alireza Ghafarollahi2 and Markus J. Buehler2-4
1Department of Chemical and Biomolecular Engineering, University of Illinois at Urbana-Champaign
2Department of Mechanical Engineering, Massachusetts Institute of Technology
3Department of Civil and Environmental Engineering, Massachusetts Institute of Technology
4Schwarzman College of Computing, Massachusetts Institute of Technology

2D materials such as graphene, first isolated in 2004 by Andre Geim and Konstantin Novoselov, have
the potential to lead to revolutionary technologies in electronics, energy storage, and manufacturing
due to their unique mechanical and electrical properties. However, the design space of these materials
is essentially infinite, making it unfeasible for humans to search manually. Machine Learning tools
are frequently used for exploration and optimization of properties in materials, but they rely
heavily on curated databases with multiple known variables. General-purpose Large Language
Models (LLMs) like GPT-5.6 or Fable, on the other hand, can fill in gaps in information through
their broad knowledge bases. Hence, autonomous AI frameworks, when combined with evolutionary
algorithms and first-principles calculations, can generate creative structures with potentially useful
properties. Here, we present Sparks2D, an autonomous scientific discovery system, to uncover
mechanistic principles in graphene. To test a proposed hypothesis, AI agents propose different
designs of modified graphene, which are evaluated through physics-based simulations. Sparks2D
then iterated upon designs through a “survival of the fittest” evolutionary process, based on a scoring
system that prioritizes optimization of target material properties. Through this, I aim to enhance the
self-directed innovation of Sparks2D for inverse design of materials.