I mentioned in the sync that I did a little Parquet file metadata
visualizer.
I wanted to have something to help people who are benchmarking Parquet and
want to take a look at what the pages look like and verify that the choices
on page size or encoding are not creating degraded performance.
This is made with claude code and doubles as my personal experimentation
with AI coding.
Note that I don't really know typescript or front end development, so code
quality may vary.
Requirements:
- run entirely in the browser. (It is using hyparquet!)
- display the page metadata by iterating through the page headers for a
given column chunk.
- it is trying to be careful of memory usage, I have loaded 2GB+ Parquet
files with many columns in it.
https://github.com/julienledem/parquet-metadata-vizualizer
I have published it here, if you want to play with it:
https://julien.ledem.net/experiment/parquet-visualizer.html
There is no backend. It is all running in your browser.
Here is an example with the NY taxi data
[image: Screenshot 2025-10-30 at 10.58.53 AM.png]
