I'm in my 3rd year at Berkeley studying EECS and am excited by BioML, language model reasoning and interpretability. I currently work on viral protein language models with the Marks Lab and metagenomics analysis with the Wolf Lab.
I lead the research commitee in Machine Learning at Berkeley where we mentor homegrown research projects, discuss papers and support members with funding to attend conferences. I also co-organize the BioML Seminar Series with Aakarsh Vermani, bringing leaders at the cutting-edge of comptuation and biology to Berkeley.
I post-trained Gemma 3 1B with Tunix (JAX) using GRPO to make outputs reliably follow a structured format. Trained on math (GSM8K, SVAMP, MultiArith) and QA (SQuAD v1, Natural Questions), improving both format compliance and answer correctness.
A comparison of receiver heads in Gemma-3-4B-IT and Qwen3-4B-Thinking.
Implemented and benchmarked beam search with multiple penalty schemes (n-gram, Hamming, cumulative) and Monte Carlo Tree Search (MCTS), evaluating accuracy on MATH Level 5 problems.
Sequence Alignment is Everything!
