• PE is available as part of db as well, like swtable, topk, count, notification etc
  • Interface for PE defines minimum set of public functions
  • There would be different implementations of the interface for different algo /sdk etc
  • There would be PEManager, who would create these individual PE
  • PEManager would store the metadata related to PE in PE table in BangDB
  • The output of PE may be stored in DB based on stream definition, PEManager to do the same
Deep Learning Diagram