Research Projects

Optimization of Social Media Marketing Strategy

Rice University |  Rice D2K Data Science Capstone (in collaboration with Lineups.io)  |  Aug 2026 – Present

• Collaborate with an interdisciplinary team to build an end-to-end sequential pipeline for Lineups, reframing digital engagement into an optimal stochastic control framework.

• Formulate a POMDP framework coupled with Hidden Markov Models to estimate latent audience states and derive optimal posting policies under sparse observational data.

A Cost Benefit Analysis of Pulling Starting Pitchers in the MLB

Rice University |  Indergraduate Research in Statistics  |  Advisor: Jeff Dominitz  |  Summer 2026

• Built a Monte Carlo simulation engine parameterized by empirical times-through-the-order degradation curves to correct for structural censoring.

• Evaluated managerial decision-making to identify counterfactual, run-minimizing starter removal thresholds.

• Links: [Working Paper] [GitHub]

Quantifying the Disruption Tax: A Run-Value Framework for Pitcher Influence

Driveline Baseball  |  Summer 2026

• Developed a custom run-value framework measuring the degree to which pitcher timing variations and delivery decoupling disrupt hitter synchronization.

• Decomposed continuous tracking and biomechanical markers to quantify the marginal run-prevention value of deceptive delivery tempos.

• Links: [Working Paper] [GitHub]

The Swiss Army Swing: Swing Repertoires, Adjustability, and Their Value in MLB

Driveline Baseball  |  Summer 2026

• Built an unsupervised Gaussian Mixture Model (GMM) framework to classify MLB swing profiles and value situational adjustments.

• Isolated intentional mechanical adaptations from spatial pitch location via orthogonal residualization to benchmark counterfactual swing decisions.

• Poster Presentation: Carnegie Mellon Sports Analytics Conference (CMSAC) (Oct 2026).

• Links: [CMSAC] [Abstract] [GitHub]

Introducing Blend+: Evaluating Pitch Arsenals as an Orchestra

Rice University  |  Sport Analytics Seminar (SMGT 490)  |  Jan 2026 – Sep 2026

• Built a hierarchical XGBoost framework and the "ArsenalFX" metric to isolate pitch repertoire synergy and expectation distortion from isolated physical pitch movement.

• Oral Presentations: 

  • Cascadia Symposium on Statistics in Sport (CASSIS), Vancouver, BC (Sep 2026)

  • Sabermetrics, Scouting, and the Science of Baseball (Saberseminar), Chicago, IL (Aug 2026)

• Poster Presentation: SABR 54 National Convention (Yoseloff Scholar) (July 2026)

• Links: [CASSIS Slides] [SABR Poster] 

Arsenal+: Quantifying the Effects of Pitch Arsenal Diversity

Rice University  |  Spring 2026

• Built a multi-class XGBoost pitch outcome model pipeline to isolate ArsenalFX—quantifying the marginal performance value of 3D pitch-plane separation and repertoire contrast.

• Featured Presenter: ASA Virtual Sports Analytics Conference (Rising Stars) (Apr 2026) 

• Awarded Best Online Presentation: Ohio State Sports Analytics Conference (Mar 2026)

Closing the Gap: Measuring Defensive Back Pursuit Efficiency Using Tracking Data

NFL Big Data Bowl Submission  |  Fall 2025

• Co-engineered an Expected Closing Time model utilizing high-frequency (x,y,t) coordinates to quantify defensive back recovery paths and spatial pursuit geometry.

• Modeled ball-carrier and defender dynamic interaction effects using orientation vectors and instantaneous velocity to isolate marginal gains in explosive play prevention.

• Links: [Kaggle Submission Writeup] [Code]

A Bayesian Exploration of Batter Performance in High Pressure Situations

Rice University  |  Advanced Sports Analytics Final Project  |  Fall 2025

• Applied Bayesian Hierarchical Modeling and empirical shrinkage estimators across 5 seasons of Statcast data to isolate latent batter talent from small-sample situational noise.

• Developed a custom Markov Chain-based Pressure Index using leverage matrices to mathematically define and quantify high-leverage contexts.

• Links: [Paper PDF] [R Code]

Identifying Qualities of Successful NECBL Pitchers Using Stuff+

North Adams SteepleCats(NECBL)  |  Summer 2025

• Architected a Stuff+ and Location+ framework using XGBoost models on 98,000+ TrackMan deliveries to predict run values derived from empirical linear weights.

• Utilized K-Means clustering and SHAP analysis to quantify how Vertical Approach Angle (VAAvsExp) and arm-slot deviations drive deception beyond raw physical traits.

• Links: [Research Paper]