I build tools, forecasts, and models that make complex ideas clear and actionable.
- Now
- Founder, Name of Record · independent data & web contractor
- Education
- M.S. Computer Science (AI), Dartmouth, 2026 · B.A. Mathematics & Government, 2024
- Known for
- Solved Erdős Problem #659 (1997), accepted by Terence Tao (arXiv)
Selected Work
Solution to a Problem of Erdős
Solved Erdős Problem #659 (1997), accepted by Terence Tao on the Erdős Problems forum. Constructed n-point planar sets with O(n/√log n) distinct distances under 4-point constraints using a human-AI proof workflow.
Combinatorics · arXiv · Forum thread · Dartmouth CS
2024 Presidential Election Forecast
Built from a weighted average of election polls, with polling-error distributions estimated from backdated election data and outcomes simulated by Monte Carlo. Provided state forecasts, historical trends, and polling analysis to thousands of viewers throughout the election.
Statistics · R · Interactive forecast · Methodology
Evaluating Agentic Research: Single-Agent vs Autonomous Multi-Agent Decomposition
A large-scale empirical evaluation of single-agent vs. autonomous multi-agent LLM deep research systems, showing no accuracy gains but ~2× higher cost and slower execution for autonomous decomposition.
LLM evaluation · Multi-agent systems · Preprint
State Media Tagging Does Not Affect Perceived Tweet Accuracy
Co-authored a preregistered U.S. survey experiment (N≈2,500) finding state-affiliated media labels had no effect on belief accuracy, while fact-check labels significantly reduced misperceptions.
Survey experiment · HKS Misinformation Review
LLM Multi-Agent Code Review
Designed and benchmarked an agentic code-generation framework (MCP, LangGraph, GitHub) on HumanEval, showing single-agent ReAct outperformed two-step orchestration by ~11%, with failures driven primarily by tool-use errors rather than code correctness.
Agentic systems · Poster
Transparent Activation Steering
Developed and evaluated an activation steering method for LLM toxicity control by training linear probes on mid-layer activations in Gemma, achieving significant manipulation of toxicity levels on in-distribution prompts.
Interpretability
Lucid Dreaming Induction System
Collaborated on a real-time EEG/EOG neurofeedback system with overnight deployment for lucidity induction; designed and implemented eye-movement (LRLR and REM) classification models used for low-latency sleep-stage inference.
Neurotech
3D Particle-Based Fluid Simulation
A discretized 3D particle fluid simulation based on the Lagrangian formulation of incompressible Navier-Stokes, with solid body coupling, irregular containment, and interbody friction.
Physical simulation
Bad Actors in Viral Fact-Checking
A computational model of viral belief spread, showing that even a single immutable adversary can enable endemic persistence of hoaxes in ER and BA networks.
Network science · Interactive model
Experience
Name of Record
Founded Name of Record, a service that maintains canonical, fact-checked biographical records.
Founder · nameofrecord.com
Lighthaven
Completed a monthlong writing residency and fellowship at Inkhaven in Berkeley, CA.
Inkhaven Resident
Independent Contractor
Multiple small engagements, including contract-based data analysis in R (data wrangling, analysis, and visualization) and the development, deployment, and maintenance of lightweight web applications.
Data & Web
Farallon Capital Management
Developed and deployed a full stack pipeline for prediction market data (Kalshi and Polymarket). Data pulled from APIs on a daily cycle, processed into a database, and visualized on information pages and in a custom search engine.
EXPLOR Intern
CND Life Sciences
Developed and deployed an interactive web application to help the sales team interface with customer data, with over 50 hours of engagement in its first 100 days. Presented insights to company leadership that led to policy change.
Financial Analyst Intern
Dartmouth College
Compiled datasets of responses to digital anti-extremism campaigns and conducted sentiment analysis on textual data through an LLM wrapper.
Research Assistant
Writing
On my Substack I write about AI and politics, mostly independent of each other. Formerly the home of my election forecast.
About
I enjoy unraveling the mechanisms of the world we live in, from incompressible fluid dynamics to the network science of viral information. This drive led me to study mathematics, government, and computer science while exploring topics as diverse as psychology, physics, and philosophy. More recently, I've become deeply engaged with the field of AI.
At the London School of Economics I researched emerging technologies in interstate gray-zone conflict. I attended Y Combinator's 2025 AI Startup School. Other interests include mountaineering, wilderness first aid, professional cooking, and Tae Kwon Do (second degree black belt).