Projects

Here are some of the most exciting projects I’ve worked on — combining machine learning, NLP, time series forecasting, and music AI.

🏠 Colorado Housing & Rent Forecasting

Tools: Python, scikit-learn, statsmodels, Elastic Net, XGBoost, SARIMAX
Description:
Ongoing time series modeling project to forecast rent and mortgage trends for Colorado single-family homes using Zillow and economic indicators. Applied transformations, lag features, and regression diagnostics.
Key highlights:

• R² up to 0.97 on validation
• Elastic Net and XGBoost models for multicollinearity and performance
• Time-series aware training/test split
• Interpretable with SHAP and feature correlation
• Final code in progress Github • Read Report

🧠 Multimodal Interview Scoring with NLP

Tools: TF-IDF, Word2Vec, prosodic features, Random Forest, FNN, SHAP
Description:
Developed a multimodal machine learning pipeline to predict excitement and performance scores from interviews using both textual (NLP) and prosodic (audio) features. Applied explainable ML techniques to interpret model predictions.
Key highlights:

• 5-fold cross-validation using TF-IDF and Word2Vec
• Multimodal models outperformed unimodal baselines
• SHAP and correlation analysis for interpretability
• Based on the MIT Interview dataset
• Final report and results complete Github • Read Report

📰 Fake News Detection with NLP (In Progress)

Tools: TF-IDF, Logistic Regression, XGBoost, BERT (in progress)
Description:
A binary fake news classification project using the LIAR dataset (12K+ political statements). Combines traditional NLP (TF-IDF, Word2Vec) with machine learning models like Logistic Regression and XGBoost. Currently experimenting with fine-tuning BERT for improved performance.

Key Highlights:

• Mapped 6-class labels to binary: Real vs. Fake
• Achieved ~62% accuracy with Logistic Regression and XGBoost
• Evaluated using precision, recall, F1, and confusion matrices
• BERT fine-tuning and web demo deployment in progress

GitHub Repository

🎼 Wolfie: AI-Powered Mood-Based Music Generator (In Progress)

Tools: Python, PrettyMIDI, FluidSynth, Symbolic AI, Music Theory
Description:
An original music generation system that composes emotionally expressive chord progressions and melodies based on user-selected moods and classical harmony logic. Wolfie blends symbolic AI and music theory with the foundations of generative models.

Key highlights:

• Rule-based chord generation (Circle of Fifths, cadences, key modes)
• Melody creation with customizable inversion, tempo, and emotion
• Outputs expressive MIDI compositions
• Inspired by Bach, Mozart, and Beethoven
• Currently integrating LSTMs and deep learning for expressive control

GitHub • Demo coming soon