In psychophysical vision experiments, patients are shown a stimulus on a screen, and are asked to respond. An essential source of data collected is the movement of the patient's eye in response to the stimuli, which is recorded as a video by a Scanning Laser Ophthalmoscope (SLO). I developed open-source, modularized, and also standalone software to computational analyze this retinal imaging data, which we call ReVAS.

When a neuron is stimulated by an electrical signal, a voltage response is emitted. An accurate model of this gives us a cheaper way of performing experiements, and a better biophysical understanding of the system. By compartmentalizing our model and assigning parameter values, we can compare empirical voltage responses to our generated predictions according to a learned linear combination of scoring functions and an analysis of the sensitivity of each parameter. I architected a pipeline for running optimizations, visualizing results, and organizing our vast quantities of datasets through a self-documenting format called NWB (which is based upon HDF5).

When antibiotics are used to combat bacteria, a full course must be delivered in order to ensure that any uneliminated bacteria do not have the opportunity to replicate and allow its population to evolve resistance. Failure to avoid imposing naturally selective pressure can have catastrophic public health consequences, as bacteria can evolve more rapidly than we can produce new drugs. I analyzed the popular and scientific press’s use of evolutionary terms to understand how this phenomenon is presented.

I taught the "Data Structures" course at UC Berkeley as a Teaching Assistant for 5 semesters, and in Summer 2019, I was the lecturer for an official offering of the course for 331 students. During the summer, I took the lead on authoring the exams, delivered custom lectures, and revamped the lab assignments linked below.

We generate commit messages from git diff logs, comparing the performance of state-of-the-art language models including recurrent neural nets with LSTMs, the Transformer, BERT, and GPT-2.

We apply machine learning techniques and models to determine the pH of chemical solutions given raw image data of pH test strips. We report our findings in comparison with those in the literature.