Investigated RNN, transformer, and diffusion-based neural decoding models for reconstructing stimulus images from peripheral retinal ganglion cell (RGC) spike trains in macaque monkeys.

Designed a transformer encoder-decoder as a neural encoding model for predicting spike trains from stimulus images.

Advised by Professor E.J. Chichilnisky via the Stanford Artificial Retina Project.

Wrote end-to-end pipeline for downloading and processing transmission acknowledgement files received from banks that are involved in thousands of critical money transfers daily, using Java and MySQL.

Developed receipt scanning functionality using Apple's Vision framework as part of a team of 4 working on a bill splitting feature within Cash App that I proposed for intern hack week.

Won intern hack week and pitched a mockup to Square's core executives and CEO Jack Dorsey.

Improved Precision@1 by 9.1% for feedforward model that predicts shopping category from an input query.

Updated team's Docker image used in offline data building pipeline, powering 5 of the team's deep learning models to support Python 3.6 and TensorFlow 2.2.

Redirected AWS Batch job logs to enable more convenient debugging support for team using AWS Batch/Logs API.

Built a deep learning model that classifies the emotion of an utterance. Developed a novel feature extraction method that combines handcrafted and learned features. Our results beat the state-of-the-art accuracies in speech emotion recognition (SER) for speaker-dependent experiments on the IEMOCAP dataset by 6%.

Gained proficiency in PyTorch for writing deep learning architectures; NumPy, SciPy, and Pandas for data preprocessing; and Matplotlib for data visualizations.

Read more here: https://rpc21.github.io/data-plus-results/

Paper featured in the 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP): https://ieeexplore.ieee.org/document/9054629

Research done through the Data+ summer program at Duke.