Sam has done some amazing things with Venus, turning it into a feed processing platform. The really amazing part is at the bottom of the architecture document. Everything, absolutely everything, is turned into Atom, and not just Atom, but Atom with title, summary and content in XHTML. Let that sink in.
Once it does then you then get to the filters:
Furthermore, the design is that both filters and templates read from stdin and produce output using stdout. This means that any programming language may be used. Furthermore, as filters can be real programs, they need not limit themselves to filtering. In Unix terms, the can be tees. They can scan the input for interesting data, and POST ones that are of interest elsewhere. They can index the data using something like Lucene.
I could put Bayesian filters in place and populate a new feed using the Atom Publishing Protocol from a filter. I could filter a bunch of feeds and build up my own Statistically Improbable Phrases and use that to highlight entries. And don't even get me started on what I could with microformats...