I use a basic JSON data structure with some Go code that reads it, filters on various things (primarily date ranges at the moment), and outputs it via a template to HTML.

For multi descriptions, I'd change the description key from a string to a hash and then have an option to select those of a particular "flavour".

Something like:

    d: {"ml":"I made funny carrot names with ML", "data":"Extensive processing of botanical nomenclature", "mba":"Some guff about something business processes"}

And then something like:

    ./gencv -flavour ml > ml.html # uses ml.tmpl
    ./gencv -flavour data > data.html # uses data.tmpl
    #etcetcetc

You could even lash it up in shell with `jq` and `sed`...