Portfolio

Leonidas Varveropoulos

Software / Machine Learning Engineer

Hi, I’m a Senior in the Honors Turing Scholars Program at the University of Texas at Austin, studying Computer Science with a minor in Statistics and Data Science.

I’m passionate about designing and building creative solutions at the intersection of robotics, computer graphics, biotechnology, and process automation. With a strong foundation in software design, statistics, and machine learning, I enjoy the full design process: turning ambitious ideas into working systems.

STARTING OUT

My work in engineering began 7+ years ago with a makeshift Roomba powered by the ROS ecosystem. Outfitting it with stereo cameras, optical flow sensors, and LIDAR to implement Kalman and particle filters showed me the difficulty of getting systems to work in unconstrained environments. This challenge became a central theme of my work, and I have spent the years since exploring the technologies required to solve it.

Throughout high school, I built on this foundation through FRC Team 624, where I served as President and Lead Programmer for a program that made multiple appearances at the World Championship. Around the same time, I founded vrobotsim.com, an open-source robotic simulator designed to run entirely in the cloud. By streaming video directly to the browser, the platform allows students to experiment with robotics without any local installation. The project is still in active development today and has reached 32,000+ users.

EDUCATION

I am currently a student in the Turing Scholars Honors Program at the University of Texas at Austin, where I am majoring in Computer Science with a minor in Statistics and Data Science. This program has allowed me to solidify my background in algorithms, machine learning, and systems design through both rigorous coursework and hands-on research.

Generative Visual ComputingCloud ComputingMachine Learning I + II
Operating SystemsArtificial IntelligenceComputer Graphics
Stochastic ProcessesAutomated Plan for RobotsSpeech/Audio Processing
Notable Coursework

RESEARCH AND PROJECTS

Outside of the classroom, I am involved in reinforcement learning research at UT’s Robot Perception and Learning Lab. I am currently helping design an agent that can “learn how to learn” by adapting to play diverse retro games on the fly with a single trained model. This work forms the basis of my Turing honors thesis, which focuses on the role of world models in meta-reinforcement learning.

I recently finished Project Skydrop, which utilized sequential estimation and geospatial modeling to solve a real-world weather based treasure hunt.

Currently, I’m also exploring how video diffusion and Gaussian splatting can be combined to generate 3D models directly from video for use in simulation.

Example of predictions of the hidden location over time for Project Skydrop.

HOBBIES

Outside of computer science, I enjoy building and flying FPV quadcopters—both for long-range aerial photography and high-speed racing. I’ve also experimented with CAD, 3D printing, resin printing, and laser cutting on various side projects, including designing my own hand-stitched custom soccer ball.

At UT, I’ve stayed active through clubs and rec leagues, joining soccer, pickleball, Brazilian Jiu-Jitsu, UT Rugby, and the Longhorn Neurotech club to explore and work on machine learning models for neural signal processing.

Recent Experiences

Garmin

Software Automation Intern

Machine Learning, Computer Graphics, Digital Twin Simulation, Robotics, Distributed Control Systems

UT Robot Perception and Learning Lab

Undergraduate Researcher

Meta Reinforcement Learning, World Models, Model Distillation, Curriculum Learning

Roboligent

Machine Learning Intern

Robotics, Imitation Learning, Quantization for Realtime Inference, Fleet Management

UT Human Enabled Robotics Lab

Undergraduate Researcher

Medical Robotics, Computer Vision, Haptic Feedback, Machine Learning

See more

Featured Projects

Project

Bayesian Neural Network Treasure Hunt Predictor

This project addresses the complex geospatial modeling challenge presented by Project Skydrop, a real-world treasure hunt requiring the discovery of a hidden prize using clues. By leveraging data clues taken from the hidden location and conditioning global context from satellites, we reason about the relative likelihood of the treasure’s position across a shrinking search grid. We formulate this search as a sequential estimation problem, utilizing a Hidden Markov Model (HMM) to iteratively update our belief state. The technical…

by Leonidas Varveropoulos
Project

Neural Network Path-traced Anti-aliasing and Noise Reduction

Rendering realistic images is always a balance between accuracy and speed. If you’ve ever played a video game after seeing a rendered cinematic trailer and left less than impressed, you’ve seen this problem firsthand. This project is about tackling one of the big bottlenecks: getting clean, sharp images from path tracing without paying the huge cost of supersampling.

The idea is to use neural networks to help with anti-aliasing and denoising

by Leonidas Varveropoulos
Project

Distributed Cloud Computing for Machine Learning

The goal of this project is to understand how modern deep learning frameworks scale training across multiple devices. Rather than treating distributed training as a black box, this project focuses on the underlying principles that make large-scale learning possible, such as parallelism, communication, and synchronization. To explore these ideas in practice, I implement a custom distributed machine learning algorithm using the Spark framework, applying the same core concepts that appear in distributed ML systems within a more general-purpose…

by Leonidas Varveropoulos

See more projects