Architecturally, it sounds remarkably similar to commercial social media monitoring platforms - not too surprising, since both are essentially about watching and searching the behavior of people around certain topics/groups/keywords.
Queries ('selectors') go in one end, are presumably translated into appropriate queries at each of the external 'data sources' (best-effort translation of the original selectors into whatever the source supports query-wise) and then the results are either alerted on in real-time (surveillance) or kept longer-term (stored comms).
Content returned varies on what the provider can support.
Finally there is a search interface on top (although it looks very basic in this case - simple boolean AND/OR) to provide historic search over the data collected.
Queries ('selectors') go in one end, are presumably translated into appropriate queries at each of the external 'data sources' (best-effort translation of the original selectors into whatever the source supports query-wise) and then the results are either alerted on in real-time (surveillance) or kept longer-term (stored comms).
Content returned varies on what the provider can support.
Finally there is a search interface on top (although it looks very basic in this case - simple boolean AND/OR) to provide historic search over the data collected.