Yes, it must be an actual tech to run ML workloads inside the Postgres instance, where all the data is readily available.