I explored equity momentum strategies on U.S. stocks, starting from naive daily lookbacks and iteratively refining the approach with weekly filtering, regime filters, and volatility scaling. The journey showed how risk management improves Sharpe ratio and reduces drawdowns, while also revealing the limits of momentum in large-cap equities post-2010.

This was my final year undergraduate project, where I built a fully virtualized 5G testbed to analyze CPU-level energy consumption under varying network conditions and predict usage using machine learning models. There is no public repository available, but if you're interested in this work, feel free to reach out to me directly!

I went through the full data science pipeline to test a deliberately far-fetched question: can Spotify weekly audio trends predict the S&P 500? I merged weekly aggregated Spotify audio features (like danceability, energy, and valence) with S&P 500 weekly returns and experimented with several models—from Random Forests to Neural Networks, lagged features, and even LSTM sequence models. While the models overfit training data, none showed meaningful predictive power on the test set (negative R² across the board). This wasn’t about finding causation, but rather about practicing end-to-end data preparation, feature engineering, baseline modeling, and sequence learning. The key takeaway? Not all datasets have meaningful signals, but the process of exploring them is still valuable for sharpening your ML workflow skills.

I just upgraded this portfolio website to be fully mobile-friendly and dark mode compatible! Using Tailwind’s responsive utilities, I redesigned the layout so that every page—including Projects, Blog, and the CV timeline—looks great on any screen size. The navigation now has a clean mobile hamburger menu, blog cards collapse gracefully on smaller screens, and project modals scale perfectly without breaking. On top of that, I added seamless dark mode support by leveraging Tailwind's dark variant classes—so the entire site automatically adapts based on the user’s system theme. This upgrade improves both accessibility and overall user experience, making the portfolio feel modern and polished across all devices.

I created a fun data storytelling project exploring completely absurd correlations with the S&P 500. Instead of traditional finance signals, I compared yearly stock market performance with unrelated factors like Spotify song danceability, Formula 1 champion points, global temperature anomalies, and Premier League champion points. After visualizing these trends together in a multi-series Plotly chart, I trained a simple Linear Regression model—which surprisingly achieved a MAE of just 27.6 points! Of course, the feature importance was pure nonsense, with global temperature anomalies appearing as the 'strongest predictor.' This project highlights how easily spurious correlations can seem convincing, and why context matters when interpreting data.