I mean... maybe in this case? I feel like profile/picture based matchmaking is something an ML model could be pretty good at in theory. Match people based on physical preferences and attractiveness (get head scans and frontal & profile full body shots), basic demographic/location/financial info, fill out a questionnaire with hobbies, political views, sexual preferences, etc.
Do that for groups of satisfied pre-existing couples first to train the model on, then continue training the model on the successful matches from the app. Have it spit out X number of matches that have the highest ratings for all users, limit it to X matches per time period to limit "swiping" behaviors, then let users talk/date and provide feedback to the app about what they did/didn't like.
Obviously, it would need major privacy protections given how sensitive the info is, but that'd be a way better system than Tinder and the like. Like a super powered robo matchmaker serving up the highest probability matches.