Projects

Market Moves: Portfolio Optimization with Reinforcement Learning

Implemented reinforcement learning models to simulate trading decisions in a portfolio setting. Compared Q-Learning, Deep Q-Networks (DQN), and Proximal Policy Optimization (PPO) on historical market data, evaluating how each approach adapted to changing market conditions and risk exposure. PPO showed more stable behavior relative to simpler baselines such as buy-and-hold.

AI Song Recommender (In Progress)

Built an AI-driven music recommender that combines clustering, ranking, and similarity-based retrieval using Spotify audio features such as tempo, energy, danceability, and valence. The system groups songs into natural “mood” clusters, ranks recommendations using cosine similarity, and supports query-based inputs like mood or vibe to return personalized song suggestions.

Fraud Detection Analysis (SQL)

Analyzed transactional banking data using SQL to identify potential fraud patterns. Explored behavior across transaction amounts, timing, locations, devices, IP addresses, and login activity, with device and IP switching emerging as the strongest indicators of risk.

Market Mood Analysis (NLP / Econometrics)

Analyzed whether the tone of Federal Reserve (FOMC) statements influences short-term market movements. Applied financial sentiment dictionaries and FinBERT to 20+ years of Fed communications and tested sentiment against equity and FX reactions, finding limited predictive power and highlighting the intentional neutrality of central bank language.

AI Agent Study Assistant

Built a Python-based study tool that converts lecture notes and academic PDFs into interactive quizzes. The system extracts text from long-form documents, generates multiple question types using language models, and evaluates free-text answers using semantic similarity rather than exact matching. Deployed with a simple Streamlit interface for self-study and feedback.