Graduate Studies
University of San Diego · 2024–2026
A collection of projects completed during my graduate program. The work spans the full ML lifecycle — from exploratory modeling and computer vision to MLOps, NLP, and agentic systems. I graduated as an Alcalá 100 Honoree.
AAI-590 — Capstone
Goal: Build a temporal deep learning system to forecast multiple infectious disease outbreaks using Canadian public health surveillance data.
Outcome: Developed and evaluated multiple forecasting models across diseases including influenza and whooping cough, with geographic gap analysis validating model performance against PHAC reporting patterns.
AAI-540
Goal: Build, train, and deploy a fully-operational MLOps system predicting the impact of parking accessibility on new businesses using the Yelp Open Dataset.
Outcome: Feature store with 36 features deploys a robust model with real-time monitoring and drift feedback for continuous improvement.
AAI-521
Goal: Build and train a model to detect different species of reptiles in images.
Outcome: Model matches 531 reptile species from the BioTrove dataset at 60% accuracy.
AAI-520
Goal: Implement an agentic workflow with prompt chaining to deliver a self-improving financial news agent.
Outcome: Agent retrieves news and stock price data for publicly traded companies, with sentiment analysis of news coverage and stock price trends.
AAI-511
Goal: Implement LSTM and CNN models against a Kaggle MIDI dataset to identify four major classical composers.
Outcome: Models performed reasonably well at composer identification; scaling to additional composers remains an open challenge.
AAI-510
Goal: Leverage Bureau of Labor Statistics data to forecast future unemployment rates for the San Diego Metropolitan Statistical Area.
Outcome: DeepAR outperformed baseline models; analysis surfaced the need for additional economic and geopolitical factors in a production-grade model.
AAI-530
Goal: Design an IoT system that detects energy consumption anomalies in a smart home.
Outcome: Autoencoder-based anomaly detection paired with an LSTM forecasting model achieving 99.84% accuracy across a seven-day prediction horizon.
AAI-501
Goal: Identify and implement a model to detect emotional states in speech audio.
Outcome: CNN-LSTM Enhanced model achieved 98% accuracy; HuBERT Enhanced also exceeded 90%, validating both approaches for speech emotion recognition.
AAI-500
Goal: Identify and implement a model to aid early detection of obesity risk from lifestyle and dietary features.
Outcome: Random Forest model achieved 94% accuracy in detecting correlative effects of obesity from lifestyle factors.