DOM of a Markdown document. Because Redcarpet is strongly HTML-centric we cannot pass non-String return values around. We instead have to retain all generated s-expressions and left markers of them in the return value. The DOM is constructed back from that info after we finished rendering.

md2dom.rb

#! /your/favourite/path/to/ruby
# -*- coding: utf-8 -*-
# -*- frozen_string_literal: true -*-
# -*- warn_indent: true -*-

require 'redcarpet'

class MD2DOM < Redcarpet::Render::Base
  def preprocess str
    @sexp = []
    return str
  end

  def convert *a
    ret = [__callee__, *a]
    @sexp << ret
    return sprintf("%d\n", @sexp.size - 1)
  end

  def postprocess str
    return str.each_line.map do |l|
      n = Integer(l) rescue nil
      next l unless n
      m = @sexp[n]
      next l unless m
      type, *a = *m
      b = a.map do |i|
        case i when String then
          postprocess i
        else
          i
        end
      end
      next [type, *b]
    end
  end

  %i[
    block_code block_quote block_html footnotes footnote_def header
    hrule list list_item paragraph table table_row table_cell autolink
    codespan double_emphasis emphasis image linebreak link raw_html
    triple_emphasis strikethrough superscript underline highlight
    quote footnote_ref entity normal_text doc_header doc_footer
  ].each do |m|
    alias_method m, :convert
  end
end

usage.rb

#! /your/favourite/path/to/ruby
# -*- coding: utf-8 -*-
# -*- frozen_string_literal: true -*-
# -*- warn_indent: true -*-
require 'json'

md   = Redcarpet::Markdown.new MD2DOM
dom  = md.render ARGF.read
json = JSON.pretty_generate dom
puts json