About me
Hello there! My name is Kristen Bestavros, and I'm currently a master's student at the Center for Computing and Data Sciences at Boston University. As a data scientist, I'm passionate about using computational methods to understand people, systems, and society. My work bridges machine learning, ethics, and the social sciences, focusing on how data and algorithms shape the world we live in. My current interests are in AI safety, with a particular emphasis on red-teaming and novel safety training strategies.
As an undergraduate I studied a wide variety of data science topics and pursued a minor in statistics, so I have a strong foundational understanding of the mathematics behind the methods and tools I use. In my senior year, as part of the Kilachand Honors College curriculum, I completed my keystone project on simulating political polarization with an agent-based model. Since then I've worked on a range of projects, including data analysis for CHAPA, an affordable housing nonprofit based in Massachusetts. I've also been working on personal projects, ranging from a name anagram generator toan automated adversarial safety training loop for LLMs. You can read more about my work in the Projects section below -- you can even try out the anagram generator in your browser!
AI safety
Adversarial robustness, automated red-teaming, and novel safety training strategies.
Social simulation
Agent-based models of how beliefs form, spread, and polarize across a population.
Applied data work
Statistical and geospatial analysis for a client organization.
Experience
Fall 2025 — present
Teaching Assistant: Data, Society, and AI Ethics
BU Center for Computing and Data Sciences
- Facilitate discussion sections on fairness and bias in ML systems, value alignment, automated decision-making, and the policy dimensions of AI deployment.
- Mentor students on ethical reasoning, evaluate essays, and co-develop interactive course materials.
May 2024 — May 2025
Computational Social Science Intern
BU Center for Computing and Data Sciences · Prof. Wesley Wildman
- Developed agent-based simulations modeling how ideological beliefs form, spread, and polarize within populations, drawing on social psychology and cognitive science.
- Produced a literature review published in Acta Sociologica analyzing ideological dynamics in Norway, 2014–2023.
Sep 2021 — Aug 2022
UROP Student Researcher
BU Department of Physics · Prof. David Campbell
- Conducted computational research on nonlinear dynamical systems (the Fermi–Pasta–Ulam–Tsingou problem).
- Wrote Bash tooling for parallel computation on BU's shared computing cluster.
Education
Expected May 2027
MS, Data Science
Boston University
Deep Learning; Theory and Applications of Large Language Models; Stochastic Methods
May 2025
BS, Data Science — Minor in Statistics
Boston University, Kilachand Honors College
Machine Learning; Reinforcement Learning; Algorithms; Data Science Ethics
Projects
Adversarial LLM safety training
A co-evolutionary red-teaming pipeline for LLM safety training which tests different defender strategies and safety evaluators against evolving jailbreak attempts generated by an adaptive attacker LLM.
Anagrammer
Rearranges any phrase into plausible names using every letter exactly once. A trigram model trained on 106,000 real names, phonotactic pronounceability rules, ten name templates, and composite scoring. The full pipeline runs in your browser on the page.
The Persuasion Equation
An agent-based model simulating the interplay of personality and message factors in political polarization, built on the cognitive–motivational framework of Jost et al. Keystone project, presented at the Kilachand Keystone Symposium.
Affordable housing access
As part of a collaboration with CHAPA and BU Spark!, this group project aimed to analyze property data and provide insights about applicant demographics. My primary contribution to the project was an interactive map of affordable housing applications across Massachusetts, showing applicant flow lines, demographic breakdowns, and more.
Published work
Deciphering FPUT Metastable State: An Investigation of Resonance in the FPUT Spectra
Game Development to Gameplay: Women in the Video Game Industry
Hobbies & Interests
In my free time, I enjoy creative writing, primarily in the fantasy, magical realism, and mystery genres. I love cross-genre storytelling and nonlinear narratives, and I occasionally experiment with these in my own writing. I also love collaborative storytelling in the form of D&D, which I play regularly with my friends.
Video gaming is another hobby of mine. I play and enjoy video games of all kinds, from roguelikes to strategy games to souls-like games, but I have a particular fondness for narrative-driven open world RPGs and puzzle games. A few of my favorite games include Horizon: Zero Dawn, Dragon Age: Inquisition, and more recently, Blue Prince. The next game I play will be Clair Obscur: Expedition 33.
I love brain teasers of all kinds, and have a large collection of mechanical puzzles.