Resilient Anomaly Detection in Ocean Drifter Data (RISE @ BU)
I developed a resilient anomaly-detection pipeline for ocean drifter trajectories using NOAA Global Drifter data. First, DBSCAN clustering generates per-drifter pseudo-labels for outliers. Those labels train two models—a 1D CNN and an LSTM—to classify short windows of motion as normal or anomalous. We benchmark performance with bootstrap testing and analyze regional/year differences (e.g., West Coast vs. Gulf of Mexico; 2010 Deepwater Horizon context vs. 2024 baselines). To improve reliability and efficiency, the system also simulates Reed–Solomon–style recovery for missing segments and applies delta+gzip compression to cut storage/transmission costs.
Status: manuscript under submission (abstract only shown here).
AI for Detecting and Relieving Emotional Distress on Social Media
Built an NLP pipeline (transformer-based sentiment/context modeling) to flag early signals of emotional distress and test supportive interventions. Iterated on message content via A/B testing to improve engagement while minimizing negative responses.
ESL Reading Accelerator App
Designed an adaptive reading app that scaffolds vocabulary via contextual hints and spaced repetition. Personalized difficulty from usage analytics to improve retention and reading fluency.
Environmental Monitoring with Arduino Sensors
Prototyped a modular sensor network for air and water quality with a small dashboard and threshold-based alerts. Built for low-power deployment and multi-site scaling.