HAF writes blocks straight into PostgreSQL. You have complex data as the blockchain is working with blocks full of transactions and operations; you process them with HAF and PostgreSQL, and then you can define what you want from there and expose it as an API.

I'm just devrel versus a wizard, but a simple overview of how HAF works:

• hived runs the sql_serializer plugin. As each block is processed, the plugin pushes its transactions, operations and virtual operations into PostgreSQL.
  The plugin also reports irreversible blocks and micro-fork reverts. HAF’s hive_fork_manager then rewinds app tables as needed.
• HAF apps such as account history, Hivemind and the trackers read from that database. They don’t normally read from hived.
• block_log is an optional way to fill the node. You can replay a trusted copy, which is faster, or sync from the P2P network.