Zoe Hite

Designed and built the data layer for a Salesforce-embedded React widget that puts a patient's full care history on one chronological feed. It merges encounters, assessments, vitals, medications, goals, conditions, appointments, and tasks into one view for care coordinators.

• Wrote the design spec and implementation plan, defining a unified TimelineEvent model and a category registry, so a new data source only needs one config entry, one adapter, and one hook
• Built pure TypeScript adapters that normalize 11 Salesforce and Postgres (Prisma) API sources into timeline events, splitting lifecycle records (medication start/stop, goal set/completed) into separate events
• Implemented merge, sort, month-grouping, and time-bucket utilities with dayjs to drive the feed and an activity-density minimap
• Wrote 28 Vitest unit tests covering every adapter and utility, and designed the adapters to be framework-free so they can move to a backend endpoint without UI changes

Virginia Tech CS capstone (5-person team, ML lead): a web app for Blacksburg drivers that forecasts garage occupancy with per-garage LSTM models and recommends where to park based on your drive time.

• Trained one Keras LSTM per garage on minute-level occupancy data, using 5-step windows of occupancy, day of week, and time of day to forecast 5–30 minutes ahead, with 5-fold TimeSeriesSplit validation and early stopping
• Served the models through a Flask API that interpolates each forecast to the user's Google Maps drive time to that garage
• Weighted scoring ranks garages by predicted occupancy, travel time, and distance based on the user's chosen priority
• Express/Node backend links the React frontend (Leaflet, Google Maps Places and Directions APIs) to the ML service and a MySQL session store (Prisma, Liquibase), all packaged in Docker

Developed a virtual pathfinding lesson and interactive web tools to teach algorithms and AI concepts, improving programming education for 15+ elementary students.