The stars are only the beginning — look below, and the constellations continue.
Turn to the sun, and let the night gently slip away
|

Hi, I'm Stella (Byeol)!

I'm a 4th-year undergraduate and researcher at the University of Chicago, double majoring in mathematics and computer science. After graduation, I plan to pursue a premedical post-baccalaureate program at the University of Chicago. My academic interests include anesthesiology, complex analysis, programming semantics, quantitative biology, and comp sec/cryptography.

Currently, I'm conducting research under Dr. Haddadian at the University of Chicago, using AlphaFold to analyze the structural roles of Amyloid-\(\beta\) and \(\tau\) in Alzheimer's disease. My research also involves simulating CD4+TCR interactions in the tumor microenvironment using NAMD, GaMD, and SMD to explore their structural dynamics and therapeutic potential.

I've trained a League of Legends match outcome predictor achieving 87% accuracy and 95% AUC using just 10 champion picks — 5 from each team. I've developed a minimax-based Connect6 AI that evaluates board state and win potential using a scoring heuristic and terminal evaluation, and have also developed a self-training matchbox AI inspired by MENACE.

Outside of research, I volunteer at UChicago Medicine as part of the Wayfinding team. I help guide patients and visitors through unfamiliar spaces and answer questions. These experiences ground my technical work in empathy, reminding me that care often begins with simply showing up.

Also, if you couldn't already tell, I really like stars!

Research

Investigating A\( \beta \) Oligomerization

I am using molecular modeling to study how proteins organize, stabilize, and misfold in biological systems. My recent focus is on amyloid beta oligomerization—a process central to neurodegenerative diseases.

  • Modeling A\( \beta \) oligomers using AlphaFold 2 and 3
  • Identifying the minimum number of monomers required to form a closed, stable structure
  • Determining how many additional monomers are needed for each subsequent oligomer to also close

This work sheds light on the physical thresholds of pathological aggregation, and contributes to our understanding of how molecular size influences structural stability in early-stage amyloid formation.

View full structure →

Research conducted under Dr. Haddadian,
Haddadian Lab, University of Chicago

Simulating TCR-Peptide-MHC Interactions

I'm modeling the dynamics of T-cell receptor (TCR) binding to peptide-MHC (pMHC) complexes to better understand the physical basis of immune recognition.

  • Running GaMD and NAMD simulations on antigen-specific TCR-pMHC systems
  • Examining how TCR binding geometry alters downstream signaling potential
  • Comparing the effects of TCR structural variants in in vitro versus in vivo contexts

This work highlights how subtle conformational differences in TCRs can lead to vastly different biological outcomes, and may help explain discrepancies between engineered TCR performance in experimental systems and actual therapeutic response.

Research conducted under Dr. Haddadian,
Haddadian Lab, University of Chicago

Trust in AI Moderation

I co-authored a study titled How University Students Perceive Hate-Speech Moderation by Large Language Models, which examined how students evaluate the fairness and legitimacy of LLMs used to moderate hate speech on social media.

  • Sampled 1,000 tweets from the Davidson et al. (2017) hate speech corpus
  • Used GPT-3.5-turbo to classify tweets and estimate confidence scores
  • Surveyed 21 students on their agreement with the LLM's labels and asked them to classify tweets independently

We found widespread skepticism: participants believed LLMs exhibited unfair bias, particularly along political lines. Even when the model’s labels were accurate, concerns about fairness and transparency significantly shaped trust. Ultimately, perceived fairness had a stronger impact on trust than raw performance.

This research emphasizes the importance of transparency, accountability, and user-centered design in the deployment of AI moderation systems.

K-12 Privacy & Security

I'm conducting qualitative research under Professor Marshini Chetty examining how privacy and security policies affect K-12 students, parents, and educators.

  • Analyzing interview data from students and teachers across the U.S.
  • Identifying usability barriers in edtech privacy settings
  • Exploring perceptions of surveillance, trust, and consent in classrooms

This work contributes to the design of equitable, transparent educational technologies for children and families.

Research conducted under Professor Marshini,
Amyoli Internet Research Lab, University of Chicago

My Experience with Volunteering

What I do

At UChicago Medicine, I serve as a volunteer wayfinder, navigating patients and visitors locate their upcoming appointments or their loved ones.

  • Explaining appointment locations and next steps
  • Walking patients and families to destinations across the hospital

In a space that often feels chaotic or unfamiliar, both physically, emotionally, and/or mentally, I offer a moment of stability, a moment to think only about themselves or their loved ones.

Volunteering through UChicago Medicine's Volunteer Program

Why I Volunteer

I have always loved learning, whether it be history, math, languages, or sciences. But there was one thing I have always lacked: learning from human experiences. Human experiences are invaluable, yet I’ve often found myself confined to books—things that work only in theory, in a perfect world.

However, our world is beautifully imperfect, and it is in this imperfection that I find inspiration. There are countless things to explore and accomplish in life, but none of them are which that can be done alone.

I want to connect with people, understand their stories, learn from their experiences, and help them navigate one step of their journey.

Written in response to “Why do you want to volunteer?”

My Impact

15+

hours volunteered

221+

people helped

Last updated: June, 2025

League of Legends Match Outcome Predictor

I trained a neural network to predict the outcome of League of Legends matches using just the 10 champion picks — 5 from each team — before the game begins. The model learns synergy and counter patterns from ranked matches across Master+ games.

With over 30,000 matches of training data, it achieved 87.4% accuracy and 94.6% AUC. The input space includes champion IDs, team composition, and inferred matchup dynamics. The model was trained using TensorFlow and tested on NA and Korean Challenger SoloQ games.

Each match is represented by 29 features:

  • Champion and role encodings (20): each of the 10 picks is described by a champion ID and a role ID.
  • Team synergy (2): average synergy between all pairs on each team.
  • Team counters (2): average effectiveness of blue picks against red, and vice versa.
  • Lane counters (5): matchup scores by role — top, jungle, mid, bot, support.

Beyond prediction, I adapted the model to generate dynamic tierlists — ranking champions by their win impact within specific roles, metas, and team contexts. Instead of fixed opinions, the model reorders the draft based on shifting counter-synergies and real match data.

With no players or runes in sight, it traces the gravitational pull of the draft — and listens to the way the meta shifts.

Constellations That Shape This Patch

Loading tierlist...

Skills

Languages: Python*, Java*, OCaml, C++, C, R

Markup & Scripting: LaTeX, TCL*, Batch*, Bash*, HTML, CSS, SQL

Frameworks\( \dagger \): PyTorch, TensorFlow, XGBoost, scikit-learn, React, Next.js, Tailwind CSS

Tools: VMD*, NAMD*, GaMD*, AlphaFold*, Git, BFEE, Linux, Selenium, Wireshark, Hashcat

\( X^*, Y^* \) denotes equivalent proficiency between \( X \) and \( Y \) in each category
\( \dagger \) indicates that items are listed in no particular order; all others are sorted in descending order.