This is a really impressive technical execution. The choice to use DuckDB WASM for querying parquet files directly in the browser is clever - it avoids the need for a backend server entirely while still maintaining the performance needed for 291 million rides.

What strikes me most is how this visualizes patterns that probably weren't immediately obvious in the raw data. You can likely see seasonal usage patterns, the expansion of the network over time, and how station availability evolved. These kinds of temporal spatial visualizations are often more insightful than traditional dashboards because they make patterns immediately perceivable.

One question: with 291 million rides spanning 12 years, does the animation maintain responsiveness when rendering, or do you throttle the rendering in certain zoom levels? GPU acceleration helps but that's still a massive dataset to push through the rendering pipeline.

Great work - this is exactly the kind of thing that makes public datasets actually useful to the broader community.

This one with the main url was fine https://news.ycombinator.com/item?id=46530832