I’m a sophomore majoring in computer science at MIT. I’m interested in AI safety, formal verification, combinatorics, and graph theory.
At MIT CSAIL, in Prof. John Guttag’s group, I co-developed a test-time method that makes LLM explanations more faithful; it needs no retraining or access to the model’s weights. Before that, under Prof. Max Tegmark at the Beneficial AI Foundation, I formally verified parts of Dalek-Lite, a version of the Rust elliptic-curve library that Signal and Tor use. Using Verus, I proved some of its arithmetic functions correct, along with the number theory behind them. Around the same time, at Cybersecurity at MIT Sloan, I used graph models to study how cyber risk spreads through networks.
Starting in high school, I did network science research with Prof. Wayne Hayes at UC Irvine. I redesigned two sampling methods in BLANT, his group’s tool for analyzing large networks through their small subgraphs.
I also have a background in math olympiads and competitive programming, including three years on Bangladesh’s IMO team (bronze in 2023 and 2024). I now mentor at the camps that train the team, and I led the mentor team there in 2024 and 2025.
Outside research, I play chess: I’m an Arena FIDE Master and played on MIT’s A team in the Collegiate Chess League. I also built ssehC (“chess” spelled backwards), a chess engine based on the VICE tutorial; you can challenge it on Lichess.
Selected papers
A Removal Based Approach to Improve LLM Faithfulness at Test-Time
An LLM’s explanation can leave out what actually drove its answer. We remove the concepts the explanation doesn’t credit and ask again, so the new answer can’t depend on them.
Graphlet- and motif-based link prediction in large networks
BLANT-Predict ranks likely missing links using only a network’s structure, from the small subgraphs around each pair of nodes. On 12 networks with up to about a million nodes, it was more precise than 13 other methods.
* Equal contribution. † Equal senior authorship.
Selected awards
- International Mathematical Olympiad: two bronze medals (2023, 2024)
- Asian Pacific Mathematics Olympiad: two silver medals (2022, 2023)
- Putnam: 231st of 4,335 (2025)
- Codeforces: Master (peak rating 2147)
- ICPC Asia Dhaka Regional: 6th (2023), on a guest high school team