CoralEye ( visualisation for research)

What did you make?
I built CoralEye, an edge-AI visualization dashboard for reef health monitoring around Bali. It combines a Leaflet map with Turf.js spatial analysis to render offshore sensor grids, mask out land coordinates, and run client-side TensorFlow.js MobileNetV2 image classification to score coral bleaching stages in real time.

What was challenging?
Getting TensorFlow.js graph models to run smoothly client-side while matching the training pipeline’s pixel normalization.

Fixing a class-mapping mismatch where Keras’s alphabetical folder sorting (Bleached, Dead, Healthy) inverted the severity scale.

Writing spatial logic to filter out land-based artifacts using Bali coastline polygon intersections.

What are you proud of?
Real-time browser inference that processes reef images locally without an external backend.

Building a clean research visualizer that doubles as a functional citizen-science tool.

What should people know so they can test your project?
Click around the Bali coastline to check offshore sensor popups and hex-grid zones.

Test the community report feature by uploading an underwater photo and dropping an offshore map pin to see the model calculate health scores instantly.