DVC is saving files in the content-addressable way to be able to find the previous version of a file, directory. If you push files directly, the question is how do you find the previous state of the directory (e.g. if you need to remove some files?).

If you'd like to have files in a human-readable format, I would recommend a bit different setup / workflow.

Consider using DataChain + DVC as shown here, for example: https://github.com/shcheklein/example-datachain-dvc (+ DataChain gives a way to manage and query data granularly).

The difference for you in this case is that you push, upload, modify files directly in the cloud. DataChain captures a snapshot of a state of the bucket(s) when you need to produce a new version of a dataset and saves that snapshot into DVC.

So, instead of copying data into DVC, you are essentially saving a file with references to the original data. It's a different way of doing versioning if you wish where both tools work nicely together.