Projects & Research
A showcase of my technical endeavors, research contributions, and competitive achievements.
Citadel MathDash Collegiate Quant League
Finalist, 4th Nationally
Participated in a 6-week online quant-oriented math competition hosted by Citadel.
Key Highlights:
- Placed 4th nationally.
- Invited to the Finale in Boston.
Technologies Used:
UChicago Trading Competition
Team Captain
Led a team in the 2025 UChicago Trading Competition, which consisted of two challenging cases focused on algorithmic trading and portfolio optimization.
Key Highlights:
- Case 1: Led development of a high-frequency algorithmic trading system in Python, implementing market-making, ETF arbitrage, and speculative strategies while managing risk effectively.
- Case 2: Designed and implemented a portfolio optimization solution using quantitative models (including Markowitz mean-variance and risk parity approaches) to allocate assets based on historical intraday data, resulting in consistent risk-adjusted returns and an annualized Sharpe ratio of ~9.0.
Technologies Used:
Sports Betting Algorithm
Creator
Collaborated with a non-technical friend to design and build a system to identify and capitalize on inefficiencies in NBA first-basket parlays.
Key Highlights:
- Identified patterns of players being incorrectly assigned odds based on points per game rather than true first-basket probability.
- Built a Java-based parlay calculator to find betting inefficiencies and output optimal combinations.
- Automated wager placement via Node.js (restricted usage to test environments for legal compliance).
- Backtesting revealed positive returns.
Technologies Used:
Using Data Augmentation to Improve the Performance of the ResNet-18 Model in Identifying Tuberculosis in Chest X-rays
Machine Learning Researcher
Collaborated with Dr. Parsa Akbari (University of Cambridge, now at Regeneron Pharmaceuticals) to enhance the ResNet-18 convolutional neural network for tuberculosis detection by applying data augmentations on a public chest X-ray image dataset (T. Rahman et al. 2020).
Key Highlights:
- Achieved a six-fold improvement in training time and a 43% reduction in F-score error as compared to the control model.
- Significantly outperformed human diagnosis accuracy.
- Findings documented in a research paper and presented at the Westchester-Rockland Junior Science & Humanities Symposium (Placed 5th in Computational Biology & Bioinformatics).