Projects
How I used Data Science to help me invest better in the stock market
Can data science help you invest more intelligently in the stock market? I explored this question by building a machine-learning pipeline that predicts short-term stock movements using historical data and technical indicators. This article shares the methodology, results, and key lessons learned from applying ML to real investment decisions.
Can local economic growth be predicted by satellite images?
This article explores whether satellite imagery can be used as a proxy to predict local economic growth across U.S. counties. Using nighttime light intensity, road networks, urbanization, and climate variables, I conduct a county-level panel analysis to study their relationship with GDP growth. The results show that changes in satellite-observed signals are statistically significant predictors of local economic activity.
How I built my personalized news summarizer using relevance scoring
This article explores whether satellite imagery can be used as a proxy to predict local economic growth across U.S. counties. Using nighttime light intensity, road networks, urbanization, and climate variables, I conduct a county-level panel analysis to study their relationship with GDP growth. The results show that changes in satellite-observed signals are statistically significant predictors of local economic activity.
Creating a Personal Photo Assistant with Python
This article walks through building a personal photo assistant in Python that can automatically detect faces in images, recognize known individuals, and automate sending them their photos. The author breaks the solution into key parts: face detection using modern libraries like InsightFace, building a face embedding database, and matching faces in new photos to known identities. With this system, you can automatically identify people in your photo collection and email them the pictures they appear in, streamlining what is often a tedious manual task
Is Reinforcement Learning truly the better option than A/B testing ?
Here I investigate the fundamental differences between traditional A/B testing and reinforcement learning (RL)through a simulated experiment using a real logged bandit dataset. It explains how offline evaluation with propensity scoring allows comparison of historical A/B policies with RL policies, correcting for exposure bias and estimating counterfactual performance. The analysis shows that while A/B tests are useful for simple comparisons, a contextual RL approach can outperform static A/B baselines when personalization, exploration, and adaptive decision-making matter most. Ultimately, the article argues that RL isn’t simply “better” in all cases, but becomes essential in dynamic environments where learning from feedback leads to more effective policies.