Overview

Glacast shows glacier retreat from 1984 to today using Landsat satellite imagery, then forecasts the loss through 2050. You can explore well-known glaciers and switch between 1.5°C, 2°C and 3°C warming scenarios to see how much each one changes the future.

What I made

I built the frontend with a teammate who handled the backend and models.

• UI/UX design: Designed the interface in Figma.
• Frontend: Built the site in React and TypeScript, around the map and the satellite imagery timeline.
• API integration: Connected the frontend to our FastAPI backend, which serves the historical glacier measurements and the forecasts.

Challenges

• Data with gaps: Finding datasets that held up across decades was harder than we expected. Many sources had gaps, inconsistencies, or coverage that didn’t line up with each other.
• Scope that grew: We planned a simple slider for the imagery and ended up building a full scrubber across every year from 1984 to today, all inside the 36 hours.

Takeaways

• The gap between “data exists” and “data is usable” is enormous.
• How you show uncertainty matters as much as the number itself. Our forecasts show confidence intervals instead of a single line, so nobody reads a guess as a fact.

Result

• Shipped all three machine learning features within 36 hours.