When I'm not creating fun projects or learning AI theory, you might find me playing sports like pickleball, volleyball, and surfing, or traveling to new places. I love trying new things and learning new topics.
I'm a graduate student at UC San Diego specializing in Artificial Intelligence, with hands-on experience developing RAG systems at NASA and building high-impact ML models in the healthcare industry. My work spans computer vision, like real-time robotic tracking, and NLP with Transformers and LLMs. I'm passionate about applying state-of-the-art research to solve complex, real-world challenges and shape a future where technology improves our health, relationships, and understanding of the world.
At NASA's Ames Research Center, I architected a specialized chatbot using a LightRAG framework to query complex aerospace data. My work involved engineering a custom time-based knowledge graph to improve the RAG model's performance and precision.
In the UCSD Cognitive Robotics Laboratory, I developed and deployed a real-time, multi-modal person-tracking system for a Boston Dynamics robot, enabling it to autonomously follow users with high accuracy by fusing RGB-D and skeletal data.
I’m currently pursuing a Master’s in CS with a focus on Artificial Intelligence at UC San Diego, where I delve into exciting AI theory and applications.
At UnitedHealth Group, our team developed AI models to prevent adverse health outcomes by predicting them before they happened. My role involved leading data analysis, data collection, model development, and model deployment for multiple projects. This also included creating pipelines and parallelizing GIS data collection, which empowered us to integrate critical insights into machine learning models and predictive analytics.
I graduated summa cum laude from Gustavus Adolphus College in 2022 with a 3.96 GPA, where I laid the groundwork for my career in tech.
Applied BERTopic to extract and analyze key topics from podcast transcripts, processing over 3,000 YouTube transcripts with enhanced topic labels using GPT-4o, achieving a classification rate of 78.56% for news podcasts.
Developed multiple CNN architectures for semantic segmentation on the PASCAL VOC-2012 dataset, achieving 83.3% pixel accuracy with a pretrained ResNet-based model, significantly improving on baseline performance.
Developed a natural language query interface to interact with a digital twin, using English to query both Neo4j graph databases for structural data and InfluxDB time-series databases for sensor data.
Implemented and compared multiple Deep Q-Learning architectures for maze navigation, including CNNs, Masked Q-Learning, and Dueling DQN, achieving optimal performance with a hyperparameter-tuned network reaching 100% win rate and minimal loss on complex mazes.
Implemented a neural network from scratch using NumPy to classify clothing images, experimenting with different activation functions and regularization techniques to achieve 88% test accuracy.
Used Azure Machine Learning Studio to train and hyperparameter-tune a model to predict individuals' general health from survey data, deployed as a scalable API for quick and secure health assessments.
Classified individuals into 4 types of movie watchers to predict future reviews, accomplished through 256 iterations of Expectation Maximization on a dataset of movie reviews.
Built a program to compute unigram and bigram log likelihoods of character sequences, tokenizing by word and combining models to compute sentence log likelihood.
Implemented and analyzed attention mechanisms in a transformer, optimizing encoder and decoder patterns and evaluating different positional embedding strategies, improving classification accuracy from 33.6% to 85.86%.
Implemented and compared Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) models for character-level text generation using Shakespeare's works.
If you're interested in a deeper look at my experience, feel free to check out my full resume here.