Expanding user graphs for matches across multiple nodes isn’t just easy, it’s actually way more efficient too. Have nodes calculate match likelihood among their users and verify they’re using the same algorithm with a simple hash. Then you’ve reduced getting matches for user X from the entire population Y (calculating distance, applying filters, and then ordering by compatibility, all expensive tasks) to getting matches for user X from nodes N that are purposely geolocated nearby reducing the search space even more.

Basically not only could you do this open-source and decentralized, it would actually be more secure, safer, and way better both accuracy wise and cost wise at finding matches.