Benjamin Berol
01

Hi, I’m Ben.

I’m studying computer science at Georgia Tech, with a minor in physics. I like problems where the two overlap, especially when machine learning can help us understand or build something in the real world.

At Argonne National Laboratory, I work on reinforcement learning for molecular discovery. At Georgia Tech, I benchmark scientific computing tools and program autonomous robots with RoboJackets.

B.S. in Computer Science, Minor in Physics · May 20284.0 GPA

Looking for research and software engineering internships for summer 2027.

02

Research

Reinforcement learning
for molecular discovery

NSF REU Research Intern

I fine-tune a molecular diffusion model to generate metal-organic frameworks that could help with carbon capture and energy storage.

14→65%Successful MOF assembly
90%Less assembly runtime

I redesigned how generated candidates are evaluated and parallelized the assembly and evaluation workloads across 72 CPU cores.

View research poster (PDF)

Presenting my reinforcement learning research poster at UIUC
Presenting my research at UIUC · Summer 2026
Reinforcement learningPyTorchParallel computing

Scientific computing benchmarks

PURA Scholar & Undergraduate Researcher

I implement and benchmark sparse algorithms such as linear solvers and GCN inference on problems ranging from megabytes to terabytes to find the fastest frameworks.

3,773benchmark problems
Explore the project

I also build Array API–compliant infrastructure, automated data pipelines, and visualizations to compare runtime and coverage across NumPy, PyTorch, SciPy, and PyData Sparse. I’m writing a paper on this work as first author.

Browse the benchmark source ↗
Numerical methodsPythonSparse arrays
03

Engineering

Autonomous RoboWrestling

Software Engineer & Operator

I’m the lead programmer for our 500g and 3kg robots. I write C++ for autonomous strategy and systems integration, and use physics simulations to test robot geometries before we build them.

A match from competition.Open video ↗
Team’s software ↗
C++Autonomous systemsSimulation

ML for physics simulations

Founder & ML Developer · Klaus Startup Challenge

Through the Klaus Startup Challenge, I built Newtonic to explore how machine learning could improve physics simulations. I developed graph neural network and LLM-based tools for performance, scalability, and automated analysis.

Graph neural networksMachine learningScientific computing
04

Off the clock

I play a lot of beach volleyball, like to ski and lift, and am a huge Washington Commanders fan. I’m also always up for poker night or a concert.

With the volleyball friends I met while researching at UIUC
Volleyball at UIUCThe friends I made after lab hours.
Skiing in Park City, Utah
Park CityA few days in the mountains.
My view of Bruno Mars performing live
Bruno MarsEven better live.
↓