It's relatively low-tech, but it does work. First, I have some scripts to preprocess the Markdown and extract custom markup that I have built to do things that Pandoc's Markdown can't, but that I can still easily inline in the document. For instance, I replaced the source code markup with a custom one that generates LaTeX formatted the way I want it formatted. I use Pandoc to output LaTeX. I then have a series of shell scripts to postprocess this LaTeX to insert pre-rendered figures, re-introduce the custom markup as rendered LaTeX, or to massage the output so that the document is laid out the way I expect.

I'm planning to use this to write a book in the near future.