PROJECTS
Data science projects spanning machine learning, statistical analysis, database architecture, and software engineering.
kNN RECOMMENDER SYSTEM
Avian Species Classification
Built a k-nearest neighbors recommender system for bird species identification, featuring custom decay functions, soft-zero weighting, and multi-dimensional hyperparameter optimization. Achieved 1st place in class-wide Kaggle competition.
NEURAL NETWORK CITE
ADT Protein Prediction
Custom-built feedforward neural network predicting Antibody-Derived Tag (ADT) protein expression from RNA sequencing data. Achieved 0.858 Pearson Correlation and 1st place.
CROSS-MODAL VAE
Biological Prediction
Built a cross-modal Variational Autoencoder predicting ADT protein expression from RNA sequencing data without paired samples, using adversarial latent space alignment. Achieved 0.75 Pearson Correlation and 2nd place.
WINE AI
Predicting Tasting Notes from Climate
Predicted wine tasting note probabilities from climate data and grape varietals using a multi-model ensemble. Achieved MAE of 0.01873 and 2nd place in competition.
SVM & DIM. REDUCTION
Deconstructing Student Achievement
Coupled dimensionality reduction with a Bayesian-optimized SVM to predict high school academic success from high-dimensional, mixed-type US Census survey data. Achieved 90% recall for at-risk students.
DINNERBOT
AI-Powered Meal Planning
Serverless weekly meal planner powered by Google Gemini with a Gordon Ramsay persona. Generates dinner options, handles selections via Telegram, and produces aisle-grouped grocery lists.
LAPLACE DISTRIBUTION
Interactive Statistical Explorer
Interactive R Shiny application for exploring the Laplace (Double Exponential) distribution — featuring PDF/CDF visualization, Monte Carlo simulation with heavy-tail analysis, and real-world case studies.
GUN VIOLENCE ANALYSIS
Geospatial Intelligence
Multi-scale geospatial analysis of U.S. gun violence integrating Gun Violence Archive data with Census Bureau demographics. Revealed that poverty predicts lethality — not incidence — through Welch's t-test, permutation testing, and bootstrap inference.
ORDER HISTORY DFA
Multivariate Time Series Dimensionality Reduction
Applied Dynamic Factor Analysis via MARSS state-space modeling to reduce 39 multivariate time series variables into 6 interpretable latent trends explaining 55% of variance — then validated with a public dataset showing 11% MAE improvement over raw features.
BJJ ADCC ANALYSIS
Competition Analytics
Data-driven analysis of ADCC submission grappling championships using Tableau, uncovering winning strategies, submission patterns, and competitive trends.