generating XHTML code+RDFa dinamically

Dear Semantic Web Deployment Working Group!

I'm an Msc student and I'm planning to prepare a thesis about certain
aspects of the Semantic Web. I already had some work about the concept
considering the bottom-up approach. I intended to create a Firefox plugin
which may annotate some special existing texts on the web but challenges,
that I'm sure you're aware of seemed to be too much for my goals.

So, for now I decided to turn to the top-bottom approach.

I thought of a web application similar to Twine or Freebase where people
could annotate certain topics to eventually create a distributable knowledge
base with RDF and OWL. The site, if the thing that is being annotated
happens to be described by one of the ontology (or the users already have or
create their own ontology), could give hints for the annotator about what
properties can be annotated of that actual thing. So basically a tool that
may encourage the web user community to generate machine processable
annotations by trying to present some kind of a user friendly GUI. I know
that this is also a big task also, but the my current problem is not
exacctly this.

My basic question is more related to RDFa. If I understand correctly a new
approach is emerging which will integrate the semantics in the page (XHTML)
source code to handle human and machine 'processable' content together,
instead of storing annotations in separate files (on which my original
concept based).
Could you verify that? Shall I try to create a web application based on
RDFa, or it is enough (because possibly it is easier), if I implement it
based on the original RDF concept? If RDFa is truly the future, could you
then give me some hint there are there any tool available now or in the near
future, or are there any best-practices about how one can add these type of
semantics to XHTML dinamically? Not hard coding it into the code by a
developper, but to insert such code into an XHTML based on user annotations..
What is your general idea about is? Using a separate RDF for the annotation
as a buffer, and regenerate the XHTML from it when the annotated data is
being retrieved?

Thank you very much in advance!
/András Pongó