Experience

Multimodal Real-Time Stock Market Prediction

A real-time system that turns social-media sentiment, across X, Reddit, and TikTok, into a market signal. It uses zero-shot classification with pretrained models (no labeled data required) and a synchronized framework that aligns sentiment windows with market windows, then forecasts with time-series foundation models. Built and tested on ~188K tweets and three major NYSE indices, with multimodal (text, speech, and video) sentiment extraction.

Machine Learning for Qubit Mapping in Quantum Computing

Machine learning for quantum circuit optimization and qubit mapping. I explored reinforcement learning, graph neural networks, QAOA, and circuit cutting for noise-aware compilation, and built a PyTorch Geometric graph representation of quantum circuits that captures gate operations, qubit relationships, and topology, to support GNN and RL models that learn hardware-aware mappings with less SWAP overhead and fidelity loss.

Just-in-Time Feedback for Sport Technique with Computer Vision AI · published

A study of visual just-in-time feedback for fast-paced physical activity: a system that coaches a basketball free throw from live camera data on arm posture. In a 30-participant user study, well-timed, actionable feedback drove immediate technique correction, while some feedback types shifted how athletes saw themselves. Built during a research stay at LMU Munich, a favorite meeting point of my athletics and my research.

Read the paper ↗

Educational Data Mining on University Grading · published

A descriptive analysis of eight years of undergraduate grades at a major metropolitan university, surfacing how grading policies vary across departments, courses, and individual instructors, including a link between enrollment and average grades, and a method to flag instructors grading far from the norm. Shipped as a publicly available Python analytics tool other universities can reuse.

Read the paper ↗