CV

Argonne National Laboratory - AI Research Intern

May 2024 - Present

• Open-ended research internship studying compositional learning
• Implemented a method of fine-tuning a vision-language-action model using KTO alignment based on the method described in the paper
• Designed a data collection interface to efficiently label data in a way suitable for KTO training of the VLA model

Argonne National Laboratory - Research Assistant

September 2022 - May 2024

• Wrote the control system for a machine that uses magnetron sputtering for thin-film depositions
• Wrote data-collection software for lab equipment
• Designed and implemented data analysis software for x-ray reflectivity measurements on multilayer optics

University of Illinois - Course Assistant

January 2022 - December 2022

• Helped run a student-led course in Rust programming
• Provided guidance for students on homework and projects

University of Illinois - Software Developer

September 2024 - Present

• Worked on a team of 4 to implement analytics and data visualization features on an attendance app for a client
• Full-stack development: ReactJS frontend with modular, containerized docker backend servers

Personal Website

• Wrote a personal website from scratch with the React framework, optimized for static-loading with Gatsby
• Implemented various features that I use in my day-to-day life, such as a file hosting service
• Self-hosted with apache on a linux server

Procedura

• Designed a framework for creating procedural structures in Minecraft
• Implemented a library in Golang for creating procedural structures and communicating them to a local server
• Wrote a Java plugin that dynamically registers and loads procedural creations to provide real-time feedback

LLM Fine-Tuning

• Automated the collection of about a million chat messages from a game server I frequented, to serve as training data
• Fine-tuned Mistral 7b using LoRA adapters on a quantized base model (QLoRA)
• Training and data processing done in Pytorch, with Unsloth to accelerate training.
• Created a bot to interact with the people on the server, using the fine-tuned LLM
• Designed a data labelling interface to further refine the bot with KTO, using data from the bot’s interactions with real people.

Neural Network from Scratch

• Designed and implemented a neural network machine learning library from scratch in c++
• No 3rd party libraries used
• Efficient backpropagation and optimizer implementation

Sound Characterization Machine Learning Project

• A machine learning project to determine the pitch and timing of notes in a song that that consists of a known set of audio samples
• Based on the Basic Pitch paper from Spotify Research: (A Lightweight Instrument-Agnostic Model for Polyphonic Note Transcription and Multipitch Estimation)

Card Game Server

• Game that can be played from my personal website, featuring an obscure card game called Eleusis
• Multiplayer functionality implemented with websocket programming, and certificates updated automatically with Cron
• Frontend made with React, and backend made with Express.js (NodeJS)

SongPlayer

• A Minecraft mod that parses Midi files on the fly and converts them into a form that may be played in Minecraft
• Automatically places the blocks needed to play the song, and plays the music

HBot

• A long-running project with over 20,000 lines of code
• A Minecraft bot that aggregates many features, such as building structures, playing music, Discord integration, etc.
• Made a derivative called EpsilonBot, which is primarily focused building images out of blocks

Numerical Analysis of Projectile Trajectories in Minecraft

• Reverse-engineered Minecraft projectile physics, and produced mathematical model for all projectiles
• Uses Newton’s method to derive launch angles of projectiles with air resistance
• Created a write-up of the derivation and code on my blog

Rust Raytracer

• A multithreaded raytracing engine written in Rust
• Uses path-tracing for global illumination

Character RNN

• Character-level LSTM-based neural network inspired by an article from Andrej Karpathy: “The Unreasonable Effectiveness of Recurrent Neural Networks”
• Trained on 500 MB of text extracted from Wikipedia
• Implemented in Tensorflow
• Used in a self-hosted bot that, when given a prompt by a user, would randomly sample a completion from the predicted distribution to approximate the style of speaking of a Wikipedia article

CUDA Fractal Rendering Engine

• A configurable GPU-powered fractal animation renderer written in CUDA c++
• Produces a series of animation frames that can be compiled into a video with ffmpeg

Simple Raytracer

• A straightforward raytracing engine that supports various materials and shapes, written in Java
• Supports various material types, volumetric shading, refraction for translucent materials, reflectivity based on the fresnel equations, etc.

DLA Fractal Drawer

• A Java program that simulates diffusion-limited aggregation to produce visualizations of DLA aggregates

GLSL Shader Programs

• Various shader programs written over the course of a few years
• Primarily various ray-traced demos, ray-marched 3D fractals, and IFS fractals
• These projects are how I learned most of what I know about today about rendering techniques and 3D graphics pipelines