Research

Notes

• Random Matrix Theory Notes: Course notes for a comprehensive introduction to random matrix theory under Vadim Gorin. Topics include semicircle law, point processes, Tracy–Widom law, free probability, and universality. Available upon request.

• Infinite Iterated Function Systems: Limit sets and attractors of IIFS. Work done under Ariel Rapaport. [pdf]

Projects

A variety of projects in machine learning and computational biology.

• Model Introspection I replicated results from this paper. I LoRA post trained Qwen3 family of models to explain their responses on input ablation and activation patching experiments. High-order explainer models (those trained to explain previous explainer models), stabilize on evaluation metrics, but are worse at explaining the original model. Explainer models trained from smaller/quantized base models perform worse than those postrained from the target model on input ablation explanations, and perform similarly well on activation patching explanations in a small training run regime. Experiment yourself on the notebooks: , .

• Tumor Vasculature: In a previous life, I worked in computational immunology, using image processing tools to on spatial transcriptomics data to identify vasculature structure under Allon Wagner. It turns out this is quite hard. Contact me if you are interested in why.

• Numpy VAE: I built a classic MLP encoder/decoder VAE architecture directly in NumPy using no neural network libraries. Clone the repo to run it yourself.