Hi, I'm Jeffrey. 👋
Thanks for stopping by my page! I'm a rising junior at Harvard, studying Computer Science and Math although I have varying interests in economics, world history, and statistics.
I grew up in a suburb of Atlanta, Georgia and spent my early teens learning web development, reading up on European history, swimming laps in the pool, and enjoying the great outdoors.
I was previously an SDE Intern at AWS where I got to work on infra as well as an early engineering intern at Strala. I've also worked for companies like Scale AI and a few venture-backed startups. I enjoy designing websites in my free time.
At Harvard, I conduct research through the Kempner Institute in Dr. Mengyu Wang's AI & Robotics Lab and the Kennedy School of Government where I'm working on a cool project to streamline policy research. I'm also a managing director for Harvard Undergraduate Capital Partners, Harvard's premiere venture capital club as well as a Portfolio Manager at the Charles River Growth Fund, Harvard's oldest investment fund (est. 1994).
Where I've Been
Ongoing Technical Projects
- Benchmarking frontier model ratings of AI policy on different axes of national impact.
- Creating a web interface for interacting with medical AI agents...
- Creating a Tetris Duel Game after my favorite way to pass time.
Past Projects
FISCHER
2026 Cubist Systematic Strategies Hackathon 1st Place
A research project investigating how different meta-prompting methods (one-shot, chain-of-thought, ReAct, recursive-LM decomposition) and agentic frameworks (LangGraph specialist orchestration, multi-model judge-mediated debate, peer-vote ensembles) affect a chess engine's playing strength, search efficiency, build cost, and runtime behavior. Each of eight engines holds the task constant — build a complete UCI chess engine — while varying one axis, then is graded on the same multi-axis scorecard. The chess engine is the unit of measurement, not the endpoint.
Research Publications
Beyond Motion Primitives: Behavioral Activity Recognition from Head-Mounted IMU
Preprint · 2025
AR smart glasses need continuous behavioral context to offer proactive assistance, yet their most practical always-on sensor, the head-mounted Inertial Measurement Unit (IMU), detects only motion primitives such as walking or standing. We push beyond motion primitives to behavioral-level recognition, defining five categories that balance AR application need with sensor observability.
PaperFair Benchmarking of Emerging One-Step Generative Models Against Multistep Diffusion and Flow Models
Preprint · 2025
State-of-the-art text-to-image models produce high-quality images, but inference remains expensive as generation requires several sequential ODE or denoising steps.
Paper