The USGS created topographic maps by hand from 1884 to 2006 (when they started making their topographic map series digitally). I stumbled upon the fact that all 183,000 of their hand-drawn maps were digitized and uploaded to a public S3 bucket [0] in Cloud-Optimized GeoTIFF format [1].
Because they're in this format, it's possible to quickly mosaic images _on demand_, in a process called "dynamic tiling" [2]. The dynamic nature of it gives you a lot more flexibility, because you aren't creating static tiles once.
In contrast, the hardest part about this project is the time variation. The USGS created many different scales of maps at many different time periods, so figuring out how to combine them into a few mosaics is quite tricky. With more time invested in this, it would be ideal to have a search capability to show user-specified time horizons. (Edit: I talk about this process a bit more in the Github README [3]).
More recently I've been spending time doing similar dynamic tiling projects, but focusing on satellite imagery, to enable analytic processing of satellite imagery in the browser [4].
[0]: `aws s3 ls s3://prd-tnm/StagedProducts/Maps/HistoricalTopo/GeoTIFF/`
[2]: https://kylebarron.dev/blog/cog-mosaic/overview