In essence, it is a backtracking VM (as well explained in Regular Expression Matching: the Virtual Machine Approach) in which one of the "instructions" in the VM delegates to an inner NFA implementation (in Rust, the regex crate, though a similar approach would certainly be possible using RE2 or the Go regexp package). Then there's an analysis which decides for each subexpression whether it is "hard", or can be delegated to the NFA matcher. At the moment, it is eager, and delegates as much as possible to the NFA engine.
Small but useful: I recently discovered the thread-io crate, which makes it easy to put a reader into a background thread so that processing can happen concurrently. The win for this for me was to have one thread decoding a zstd-compressed file while the main thread parsed its contents using the csv crate. It sped things up by about 1.6x parsing a large csv file (8.7GB compressed, 79GB uncompressed) with very modest code changes.
There are some situations, when you have an existing api and someone asks for a swagger.
So opg is simple library for generating OpenAPI 3.0 docs from structs. It parses serde attributes, so you only need to specify OpenAPI related stuff.