Hi! I'm an undergraduate student at MIT pursuing Computer Science. As an undergraduate researcher at MIT's Computer Science & Artificial Intelligence Laboratory (CSAIL), I work on advancing large language model capabilities.
Exploring research in improving LLM reasoning capabilities & alignment.
We present Mixed-CoT, a training methodology that teaches language models to identify and recover from reasoning errors using error-injected chain-of-thought with GRPO optimization. Achieves 24% accuracy on perturbed problems vs 19% for standard RL, while maintaining 41% accuracy on clean problems.
Developing an adversarial self-play framework where language models generate and solve increasingly challenging coding problems. The generator creates problems that push the solver's limits while remaining solvable.
Exploring few-shot learning capabilities of LLMs on unconventional, specialized languages not in training data. We introduce a self-scaffolding approach where models iteratively add solved problems to their context window.
First ML approach to integrate radar and satellite data for deforestation detection. Our CNN + XGBoost ensemble achieves 95.08% accuracy on 134K+ geospatially aligned images.
Currently exploring research in ML!
