CV

Research Intern

2077AI

Agentic systems for few-shot robot manipulation.

• Encore: a coding agent turns three demonstrations into an executable policy program and repairs it from its own execution traces. Oral, NeurIPS 2026 Robot Learning Workshop.
• Deployed coding agents on robots to perform bimanual tasks in the real world.

Student Researcher

MIT

With Prof. Antonio Torralba, on visual generation with multimodal LLMs.

• Co-first author of Doodle Agent: Exploring Freeform Visual Generation with Multimodal LLMs (Accepted to ICCV 2025 AI4VA Workshop), a drawing agent for open-ended visual creation using LLMs and structured prompting.

Student Researcher

MIT

With Prof. Leslie Pack Kaelbling, Learning & Intelligent Systems group.

• KeyFlow (CoRL 2026 Workshop Learn@Deploy): a coding agent adapts a manipulation policy by rewriting how a few demonstrations are represented and retrieved, with no retraining.
• 62.6% mean held-out success on 13 RoboCasa tasks, against 47.8% for a VLA fine-tuned on ten times more demonstrations.

Explore HRT — Algorithm Development (Quantitative Research) track

Hudson River Trading

Invite-only program. Implemented a limit order book and an arbitrage strategy from scratch; ranked 4th of 15 by PnL.

Quant Invitational

Citadel Securities

One of 24 students invited nationwide to a quantitative trading competition.

RSI Scholar

Research Science Institute (RSI)

Authored CNOT-Optimal Circuit Synthesis under the mentorship of Henry Ma (MIT); selected as a top-5 paper out of 100.

Researcher

MIT PRIMES

Three years of mentored research on NUMA-aware concurrent data structures, alongside advanced mathematics.

B.S.

Massachusetts Institute of Technology

Mathematics and Computer Science

• Discrete Probability and Stochastic Processes (18.619)
• Fundamentals of Probability (6.7700)
• Advanced Algorithms (6.5210)
• Deep Learning (6.7960)
• Machine Learning (6.7900)
• Natural Language Processing (6.4610)
• Robotics: Science and Systems (6.4200)
• Computer Graphics (6.4400)

High School Diploma

Phillips Academy Andover

High School

MIT CSAIL Agentic AI Hackathon — Winner (Jane Street Award)

For Fireroad.ai, an AI course planner for MIT students. Award included presenting the project at Jane Street in New York.

William Lowell Putnam Mathematical Competition — Rank 254

Regeneron Science Talent Search — Top 300 Scholar

MathWorks Math Modeling Challenge — National Champion

Harvard-MIT Mathematics Tournament (HMMT) — 2nd Place, Combinatorics Round

United States of America Mathematical Olympiad (USAMO) — Honorable Mention

United States of America Computing Olympiad (USACO) — Finalist

United States of America Physics Olympiad (USAPhO) — Bronze Medalist

United States of America Junior Mathematical Olympiad (USAJMO) — Top 24

KeyFlow: Agentic Policy Adaptation with Episodic Memory

CoRL 2026 Workshop Learn@Deploy

First-listed co-first author (equal contribution with Ben Zandonati). A coding agent adapts a manipulation policy by rewriting how a few demonstrations are represented and retrieved, beating a VLA fine-tuned on ten times more data.

Generalized Planning Interfaces for Agentic Robotics

CoRL 2026 Workshop on Agentic Robotics

Second author. Ben Zandonati, Yifan Kang, Tomás Lozano-Pérez, Leslie Pack Kaelbling.

ENCORE: Few-Shot Agentic Discovery of Manipulation Strategies

NeurIPS 2026 Robot Learning Workshop (Oral)

A coding agent reads three demonstrations, writes an executable policy program, and repairs it from its own execution traces.

Interpreting Diffusion Language Models Over Denoising Time

CoLM 2026 Workshop on Non-Autoregressive Language Models

Sparse autoencoders and cross-layer transcoders for diffusion LLMs, with timestep-aware tracing and causal interventions.

Doodle Agent: Exploring Freeform Visual Generation with Multimodal LLMs

ICCV 2025 AI4VA Workshop

Co-first author. A drawing agent for open-ended visual creation using LLMs and structured prompting. Accepted to ICCV 2025 AI4VA Workshop.

CNOT-Optimal Circuit Synthesis

Research Science Institute

Authored under the mentorship of Henry Ma (MIT). Selected as the top 5 paper at RSI out of 100 candidates.