Personal Research & Engineering Projects
- Developing a mobile application for pediatric Type 1 Diabetes patients and their parents to track vitals, insulin, and daily activities.
- Building the front-end in React Native to support child-friendly, accessible logging workflows.
- Designing a Python backend integrating UM-GPT to provide LLM-powered educational support and contextual insights for families managing diabetes.
- Designed a mobile caregiving coordination app addressing gaps in multi-person scheduling, medication tracking, and multilingual communication for diverse, multigenerational households.
- Developed an accessible interface with shared task assignments, role-based permissions, health-metric tracking, and integrated translation tools to support caregivers with varying tech comfort levels.
- Built core features— vital tracking, medication tracker, calendar to reduce caregiving friction and improve adherence to daily care routines.
- Designed and engineered a context-aware, voice-activated desktop assistant that automates email composition, file navigation, and online search through natural-language intent inference.
- Implemented stateful dialogue management and semantic parsing pipelines, increasing task-completion accuracy by 35% over baseline assistants.
- Integrated a lightweight STT + intent-classification architecture that reduced response latency by 28%, improving user satisfaction in pilot testing.
- Built a real-time ASL translation and learning platform combining YOLO-based detection with a CNN classifier for robust hand-gesture recognition.
- Collected and preprocessed 2,000+ labeled images per sign, training a model that achieved 96% accuracy across 26 ASL letters.
- Developed an AngularJS + Flask web interface with real-time video streaming, enabling gesture-to-text translation under 100 ms latency.
SEC Insider Trading Analysis (Michigan Data Science Team)
- Mined 975K+ SEC filings from Q1 2020 to analyze the relationship between insider trades and subsequent market performance.
- Engineered a classification + anomaly-detection pipeline that improved significant-trade detection precision by 42% over rule-based baselines.
- Built Tableau dashboards and regression models to uncover sector-level patterns and identify temporal spikes preceding major price shifts.
Wildfire Risk Forecaster (Michigan Data Science Team)
- Developed a U-Net–based CNN to predict wildfire susceptibility from satellite imagery and meteorological data, achieving 94% accuracy on held-out regions.
- Designed geospatial preprocessing pipelines in QGIS and NumPy for seamless extraction of topographic and vegetation features.
- Collaborated with Dewberry to prepare the model for deployment in real-time monitoring; reduced false-positive rates by 22%, boosting field usability.
