I've used this approach for the last few months and I really like it. I don't see why you need to convert to HTML and then PDF though, as pandoc is flexible enough for you to write templates for each output format. My resume (https://github.com/vaibhavsagar/resume) uses that approach.
But that's exactly what my approach does though, it improves on an earlier approach which generated pdf via html by adding a ConTeXt template which allows pandoc to generate a pdf directly, and gives the resume a similar look and feel to the html version.
Using ConTeXt instead of LaTeX has the advantage that it's nicer for setting layout properties, is more of a coherent system (in contrast to LaTeX's maze of packages), and is not specifically geared towards producing scientific documents.
