I put effort into speeding up generation but not parsing. There's low hanging fruit, there. Namely, I can use a threadpool when doing the final "process sentence" pass.
Also, I'm doing regex cleanup per sentence; I feel like it's got to be faster to apply a combined regex over the whole raw text, no matter how big. That'd be a good thing to benchmark and prove.
I put effort into speeding up generation but not parsing. There's low hanging fruit, there. Namely, I can use a threadpool when doing the final "process sentence" pass.
Also, I'm doing regex cleanup per sentence; I feel like it's got to be faster to apply a combined regex over the whole raw text, no matter how big. That'd be a good thing to benchmark and prove.