Last year, I did a project with PDFs, and I just learned the PDF format, for which I am very glad.
The PDF specification is available freely, and PDF is quite accessible and easy to understand as a programmer. A PDF is a plain-text document, optionally compressed. Each page is a Cartesian plane, on which you draw geometric shapes one-by-one. It is low level and tailor-made for software-generation. Obviously there are advanced things like glyphs and things, but like any well-architected technology, you can stick to the abstraction layers if you want.
Whether to do direct PDF depends on the complexity of your documents. For basic stuff with like simple graphics, text, and images (for example, an invoice is a good candidate), then I would just write PDF directly. You'll get good experience and you will be in full control.
For more complicated things like tables and pie graphs (for which PDF is too low-level to write directly), then I would look into a library or toolkit of some kind.