It would, at scale, where there's sufficient data to train some decent embedding models to accurately judge partner compatibility. As a machine learning exercise, this is "easy". Heck, Facebook does it, possibly without even "trying" to implement dating specifically! I regularly get spammed with "suitable single women" as friend suggestions by Meta. I'm not single, but as dating suggestions go, they're actually not half bad! If I was single, I'd reach out to some of the recommendations for sure.
They've clearly implemented this exact feature, but they bootstrapped it by having a billion+ people in their system and trillions of interactions between them already. They didn't start out as a dating site.
Similarly, I've heard that LinkedIn is now more of a "single professionals dating site" than a career-making / job-seeker site, at least for a decent fraction of its userbase.
This is eminently solvable, and people do use these systems to find partners.
The nigh-insurmountable challenge is boot-strapping a dating-only site from scratch.