Piper
Fleming.

Neural News (N²)

CS 194W: Software Project, Stanford University

With Eva Geierstanger, Kenny Lam, Jack Zhang

Built an AI-powered news aggregation platform that delivers exponential news in constant reading time. A FastAPI backend scrapes and ingests articles, generates LLM summaries and tags, and serves a personalized feed with like, save, and search features. Deployed with Docker Compose and backed by PostgreSQL, with a lightweight HTML/CSS/JS frontend.

How Large Language Models Encode Demographic Identity

CS 281: Ethics of Artificial Intelligence, Stanford University

With Carolyn Hellerqvist Smith

Investigated whether LLMs systematically shift Big Five personality representations when conditioned on demographic attributes. Used the PANDORA Reddit dataset with 500 users across 4 prompt conditions (baseline, demographic hint, explicit, and combined). Found evidence of consistent demographic encoding that raises questions about fairness and identity in AI systems.

Quantum Coherence in Rare Earth Metal Complexes: A Computational Study

CHEM 161: Computational Chemistry, Stanford University

Used DFT simulations (B3LYP/6-31G basis set in Gaussian) to investigate five chemical properties governing quantum coherence in rare earth metal hydration complexes: electron shielding, electron-phonon interaction, hyperfine interactions, spin-orbit coupling, and optical coherence. Explored rare earth metals as candidate qubit materials for quantum computing, finding that coordination number influences both system size and resulting electron shielding magnitude.

Assumption Mirror

GSBGID 517: AI and Power: Five Big Questions, Stanford Graduate School of Business

With Meghna Vasudeva, Delila Kidanu

Built an interactive tool that surfaces the invisible mental model an AI builds about its user — making hidden assumptions visible through a 5-step process: asking the AI to reveal its working theory, structuring its assumptions, generating a visual representation of how it imagines you look, iterating corrections until it's accurate, then analyzing the gap between assumption and reality. Uses Claude for assumption extraction and OpenAI for image generation.

The Obesity Myth: Why Our Assumptions About Obesity Deserve a Closer Look

CS 448: Data Visualization, Stanford University

With Elsa Bosemark, Finn Staeblein

A scrollytelling data visualization that challenges common assumptions about obesity by putting them against the data. Built with D3.js, the piece walks through global obesity prevalence trends, the rise of GLP-1 medications, and the gap between public narrative and empirical evidence- guiding readers through interactive maps and charts that reframe how we think about the obesity epidemic.

Classy

CS 278: Social Computing, Stanford University

With Amanda Foess, Evy Shen, Caeley Woo

A social mobile app that turns course selection into a community activity. Users rank their classes on quantitative and qualitative dimensions (difficulty, enjoyment, overall score), follow friends to see their rankings, recommend courses, and send friend requests — all backed by Firebase for real-time sync and authentication. Built with React Native and Expo.

Learning 3D Structure in Irradiated Lithium Fluoride via Masked Autoencoders

CS 231N: Deep Learning for Computer Vision, Stanford University

With Carolyn Hellerqvist Smith

Built a self-supervised learning framework using masked autoencoders (MAE) to detect nuclear recoil-induced defect tracks in LiF crystals imaged with light-sheet fluorescence microscopy, in collaboration with the PALEOCCENE particle physics collaboration. The system tokenizes 3D microscopy volumes and trains a transformer encoder to learn spatial representations of rare particle interaction signatures- supporting automated detection of dark matter and neutrino events at scale. Initial results showed strong background reconstruction, with masking ratio sensitivity exposing the core challenge of sparse signal recovery.