I wonder how they make transactional workloads 2x faster vs normal Postgres

The analytics workloads improvements seem pretty straightforward (or at least there is prior art like timescale)

There can be improvements like what OrioleDB is trying to do: https://github.com/orioledb/orioledb/