Kairos is an AI Safety talent accelerator. As leader of and a vocal proponent of AI's risks and harms on campus, I was selected to recieve funding and mentorship to expand our presence on campus and improve research and career preparedness in the community.

I direct Tufts AI Safety, a student-run organization on campus "committed to understanding and reducing the risks of advanced AI systems". In reality, this means I run our , coordinate collaborations, research, and funding, and talk about AI issues across campus. We've successfully secured 20k+ in funding for this work and have ~40 active members.

The Lantern is the only STS-focused organization on campus. As its director I organize all of our meetings and events, coordinate with faculty and other organizations on campus, and oversee the development of our magazine. In the past we've hosted talks by Harvard's Berkman Klein Center, various cool faculty members, and community-wide discussions on the use of AI in our school.

During my semester abroad in Santiago, Chile, I built RESTful APIs to help modernize a major South American bank’s credit-card processing platform. At the end of my time there I also authored a bilingual (EN/ES) intercultural guide for the company's U.S. market goals that was reviewed the executive team.

I worked with Professor Ben Green on his book about Algorithmic Realism, providing critiques of the books theoretical claims and frameworks based on sociotechnical analyses and my knowledge of algorithmic fairness.

I was a part of the Kuali Build dev team, a business process and workflow SaaS. On the team I analyzed business workflows to deploy KB solutions, developed an internal analytics suite for c-suite reporting, and helped migrate various backend processes onto AWS (S3, Lambda).

This short piece argues that for companies to really succeed in their AI implementations, they must do so from the ground up, listening to their workers instead of imposing new tech restrictions from the top.

For my GenAI for Social Impact final project, we developed TIPS, a Chrome extension that uses an LLM to help neurodivergent users interpret tone and nuance in online text. The tool reduces interpretation time for users by 18.2%. We also ran a systematic evaluation to benchmark model bias, testing whether LLMs (GPT-4o-mini, Claude 3, LLaMA 3, etc.) alter their "bluntness" ratings of text after being told the author is autistic.

I built a collaborative filtering movie RecSys using the MovieLens100k dataset. I implemented Surprise's SVD++ algorithm to integrate implicit feedback signals, ran a pretty extensive hyperparameter search against learning rate and regularization, and then deployed an ensembly of models perturbed by +/- 10% to introduce diversity and decrease variance.

This is a brief demo policy memo I wrote about an AI-based cyber attack called "slopsquatting", which is where AI models hallucinate fake software packages that hackers can then register and fill with malware. It's framed as a recommendation to OpenAI and I outlined a few ways the organization could tackle the problem.

I built a bag-of-words logistic regression classifier with TF-IDF vectorization using scikit-learn to predict reading levels for UK English texts. The model got 2nd place out of 58 teams (~115 people).

This is an essay published in The Lantern’s Vol. 2 Spring Magazine analyzing the documentary "All Light, Everywhere". I argue that surveillance technologies like police body cameras are not objective tools, but systems whose technical design choices (field-of-view, lens distortion) inherently frame reality and reproduce bias under a false veil of truth.

This paper explores why "fair" algorithms can still perpetuate discrimination. I investigate how mathematical fairness metrics (like demographic parity and equalized odds) are fundamentally incompatible and can haphazardly reduce important sociotechnical context, using the classic COMPAS recidivism algorithm as a case study.

In this paper (written right before ChatGPT was released to the public), I discuss the AI alignment problem. I look at both the philosophical challenges of defining values and the concrete technical hurdles like reward hacking and distributional shifts, arguing that robust safety guardrails must be prioritized over rapid, unchecked progress.

At Tufts I'm mainly am a part of the Tufts AI Safety Student Association and The Lantern at Tufts. I also was on our club field hockey team for awhile.

I attented La PUC during my semester abroad with Tufts in Chile the fall of 2024.

In high school, I earned top awards at the Georgetown NAIMUN LIX & LVIII, Princeton PMUNC 2021, and All-American MUN 2019 model UN conferences, among others. I was also an Academic all-american, Captain of our varisty lacrosse team, and nationally ranked goalie.