Building a graph database is time consuming. Why are you putting edge's children (person, place, friend) in a separate table instead of using a json field? Do you plan to query the person, place, friend table directly? You need to benchmark the different queries.

My understanding is that big5 and others do not use graphdb for all their stuff since they are not as good as rdbms to do queries with a single hop or JOIN. They embrace microservices and maintain a graphdb (in memory, persistent or distributed) to answer domain specific queries. That approach is similar to your schema except that graph queries run on a single node without superfluous network roundtrips.

It would be nice to be able to use a single database for all data related stuff to have atomic writes and simpler architecture. That's what multi-model databases are tackling have a look at OrientDB and ArangoDB [1].

Also, Tinkerpop, already mentioned in the thread, is a ready-to-scale graphdb with much love I recommend you have a look at Tales of the Tinkerpop [2].

[1] https://news.ycombinator.com/item?id=10180185

[2] https://news.ycombinator.com/item?id=10316140