Generally only /companies/:companyId/buildings

And then you get a list of all buildings for this company.

Every building has a url like: /buildings/:buildingId

So you constantly get back to the root.

Only exception is generally a tenant id which goes upfront for all requests for security/scoping purposes.

This seems like a really good model. It keeps things flat and easy to extend.

I see both.

E.g. GitHub /repos/:owner/:repo/pulls/comments/:comment_id

But flat is better than nested, esp if globally unique IDs are used already (and they often are).

Yes but /comments/:comment_uuid that has a parent to /pulls/:pull_uuid is harder to map the hierarchy it belongs to.

Not really if an URL link is added to the post in the comment response.

Also it is possible to embed a sub resource (or part of it).

Think a blog post.

/blogPost/:blogPostId

You can embed a blog object with the url and title so you can show the blogpost on a page with the name of the blog in one go.

If you need more details on the blog you can request /blogs/:blogId