I'm not aware of anybody writing a Postgres backend for JanusGraph. Probably the easiest thing to do would be to study the code of the existing backends, like inmemory, berkeleyje, or cql. Beyond the backends hosted in the JanusGraph repo, you could also study the DynamoDB storage backend.

The main interfaces are:
* StoreManager keeps track of the various stores associated with a graph instance. The main store is the edgestore, but there are other stores used for ids, indexes, transaction log, system properties.
* KeyValueStore or KeyColumnValueStore. Your choice here likely depends on how you plan on storing data into Postgres. Berkeley uses KVS, while Cassandra/HBase/BigTable use KCVS.