PDFC is still in beta. If you're willing to work with prerelease code, though, it is available now.
PDFC beta tarball
The tarball contains source code, unix build files, MS Visual C++ 6.0 project files, and precompiled executables for Linux and Windows. To build under unix, run:
aclocal
autoconf
automake
./configure
make

After doing that once, subsequent builds can be performed with simply "make".
Install the program with "make install" (which must be done as root). The build produces an executable named "pdfc" and a library named "libpdfc".
To build under Windows, open the libpdfc.dsw project workspace in Visual Studio and build it. The open pdfc_cmdline.dsw and build it to get the executable command line program. For the COM wrapper, switch into the libpdfccom directory and open libpdfccom.dsw and build it. To try the program, switch into the examples directory and invoke
pdf2xml example1.xml foo.pdf
foo.pdf is a valid PDF file created from scratch. View it with Adobe's Acrobat Reader. example1.xml is intuitive to understand; have a look at it, and you'll understand most of the functionality available for creating a PDF file from scratch. Then try invoking
xml2pdf foo.pdf foo.xml
Have a look at the output xml file; as you can see, it's not as easy to understand as example1.xml. This is mostly because PDF files are inherently mostly display-oriented rather than structure-oriented; for example, there's no concept in PDF of a paragraph of text, so the best the program can do it give you individual lines as strings. However, this translation does allow you to replace text strings, images, and fonts fairly easily; replacing strings manually in PDF files is difficult because you must also update byte-counted cross reference tables at the end of the file, and replacing images and fonts is difficult for the same reason and also because descriptor data must be updated. And PDF files in general are just obnoxious to look at in a text editor.
More extensive tutorials and documentation will be made available in the coming weeks.

The program has been tested on Redhat Linux 6.1 and Windows NT 4.0. The authors would be interested to hear if there are any problems on other unix platforms. Note that the program requires the Libxml 2.0 XML parser, the ImageMagick 5.1.1 image library, and the Zlib 1.13 compression library to be installed in order to build and run properly.