Projects

A collection of my most exciting work — blending machine learning, NLP, time series forecasting, and creative AI into real-world, expressive systems.

🏠 Colorado Housing & Rent Forecasting

Tools: Python, scikit-learn, statsmodels, Elastic Net, XGBoost, SARIMAX
Description:
Developed a robust time series modeling pipeline to forecast rent and mortgage trends for Colorado single-family homes using Zillow and macroeconomic indicators. Models incorporate feature engineering, lag structure, and regression diagnostics to support practical decision-making.

Key Highlights:

• Achieved up to R² = 0.97 on validation sets
• Applied Elastic Net and XGBoost to tackle multicollinearity and improve generalization
• Engineered time-aware features and used SARIMAX for economic seasonality
• Visualized SHAP values and heatmaps for interpretability
• Fully reproducible code and documented insights
  GitHub • Read Report

🧠 Multimodal Interview Scoring with NLP

Tools: TF-IDF, Word2Vec, prosodic features, Random Forest, FNN, SHAP, EBM
Description:
Built a multimodal ML pipeline to estimate job interview performance and excitement scores using both textual and acoustic features. Combined traditional feature engineering with explainable machine learning for fairness and insight.

Key Highlights:

• Fused text (TF-IDF, Word2Vec) and prosodic features (pitch, energy)
• Performed 5-fold cross-validation with model tuning
• Achieved higher predictive performance vs unimodal baselines
• Interpreted results using SHAP and Explainable Boosting Machines (EBM)
• Dataset: MIT Interview corpus
  GitHub • Read Report

📰 Fake News Detection with NLP

Tools: BERT, TF-IDF, Logistic Regression, XGBoost, Hugging Face, Gradio
Description:
Designed and deployed a binary classification system to detect fake news using the LIAR dataset. Compared classic pipelines with transformer-based methods, fine-tuning BERT for optimal results and hosting a public demo.

Key Highlights:

• Reformatted LIAR dataset into binary: Fake vs Real
• Fine-tuned BERT model reached 76% F1 and 63% test accuracy
• Benchmarked against TF-IDF + Logistic Regression and XGBoost
• Evaluated using precision, recall, confusion matrix, and F1-score
• Live demo hosted on Hugging Face Spaces
  GitHub • Try the Demo

🎼 Wolfie: Emotion-Aware Music Generation (In Progress)

Tools: Python, PrettyMIDI, FluidSynth, symbolic AI, music theory
Description:
Wolfie is a personal project exploring the intersection of emotion, harmony, and AI. It generates expressive chord progressions and motifs based on user-selected moods using rule-based logic and symbolic musical structures.

Key Highlights:

• Harmonic engine grounded in Circle of Fifths and expressive cadences
• Emotion-driven mood → scale → chord logic
• Generates customizable MIDI compositions with tempo, contour, and mode
• Inspired by expressive works of Bach, Mozart, and Beethoven
• Ongoing expansion with LSTMs and NLP-to-music pipeline
  GitHub • Demo Coming Soon

💡 Lantern: AI Accounting Assistant (SaaS – In Development)

Tools: Python, FastAPI, QuickBooks API, OAuth2
Description:
Lantern is a secure SaaS product that connects with QuickBooks to help small businesses manage accounting tasks. The platform features modular design and secure token handling, with an NLP-based query system currently in development.

(Not Public yet – Demo coming soon)