Research

Overview

Down below are some of the research projects and groups that I have been involved. As someone who is looking to apply to grad school, being able to apply machine learning and deep learning technoques with various domain expertise is something I have always been passionate about. Currently, I am working with Professor Andrea Chiba and PhD Candidate Akshay Nagaranjan.

Future Plans

I hope to get involved with research regarding AI infrastructure, recommender systems, LLMs, and other cuttind edge applications of AI to drive change and make the world a more ethical and equitable place.

Chiba Lab @ UCSD

May 2024

  • Collaborated with Prof. Andrea Chiba to develop advanced machine listening models, creating a pipeline with Librosa and Keras to predict stress levels from classroom audio.
  • Designed a groundbreaking language model for SoundSearch that converts audio vector embeddings into rich text descriptors, integrating seamlessly with LLMs alongside PhD candidate Akshay Nagaranjan.
  • Developed a robust API for real-time voice-based stress detection, enabling transformative applications across diverse industries.
Data EngieeringMachine ListeningDL-Based Time Series ForecastingSignal ProcessingPraatParselmouthLibrosaAWS S3API DevelopmentAudio Vector EmbeddingsFeature Extractions

Mui Research Group @ ASDRP

Jan 2023

  • Developed and fine-tuned a synthetic generative pipeline using Generative Adversarial Networks and Variational Autoencoders to mitigate facial recognition interclass bias under Dr. Phil Mui.
  • Tested the augmented dataset on a pretrained CNN, achieving a significant 20% accuracy increase for minority classes.
  • Presented compelling research findings at the Southern California Conference for Undergraduate Research at Pepperdine University.
GANsVAEsCNNDeep Learning ArchitectureData Augmentation

Downing Lab @ ASDRP

Jan 2022

  • Built a chemical retrosynthesis model under Prof. Downing to predict the synthesis feasibility of various compounds.
  • Prototyped an automated chemical synthesis platform for drug discovery, integrating hardware-software interfacing with computer vision and sensor data.
Randon ForestKerasChemical RetrosynthesisCheminformaticsComputer Vision

Johnson Research Group @ ASDRP

Feb 2021

  • Developed an emotion classification model using LSTM architecture trained on data from the 2020 presidential candidates to detect biases in their policies.
  • Created an audio processing pipeline utilizing Mel-Frequency Cepstral Coefficients to extract essential audio features for model training.
LSTMPyTorchAudio ProcessingMel-Frequency Cepstral CoefficientsWeb Scraping

Subramaniam Research Group @ ASDRP

Oct 2020

  • Forecasted crime rates in Chicago using time series data with ARIMA and SARIMA models to provide data-driven insights.
  • Analyzed discrepancies in crime rates across neighborhoods, offering explanations based on socioeconomic and political factors.
  • Published research findings in the Journal of Emerging Investigators, contributing to the academic community.
Time Series ForecastingARIMASARIMAStatistical ModelingPython