diff --git a/src/marginalia/latex.clj b/src/marginalia/latex.clj
index 9aeaa89..c9cf646 100644
--- a/src/marginalia/latex.clj
+++ b/src/marginalia/latex.clj
@@ -5,3 +5,106 @@
             [clojure.java.shell :as shell]
             [clojure.string :as string])
   (:use [marginalia.html :only [slurp-resource]]))
+
+
+(def template (slurp-resource "latex.mustache"))
+
+;; ## Namespace header calculation
+
+(def headers ["subsubsection" "subsection" "section"])
+
+(def section-re #"\\((sub){0,2}section)\s*\{")
+
+(defn get-section-type
+  "It calculates the suitable header for a namespace `ns`. For example,
+  if the generated output has subsections a section must be generated
+  for each namespace in the table of contents. If it has
+  subsubsections a subsection should be generated and so on."
+  [groups]
+  (let [docs (map :doc groups)
+        matches (->> docs (mapcat (partial re-seq section-re))
+                     (map second))
+        matches (or (seq matches) ["none"])
+        most-important (apply max (map #(.indexOf headers %) matches))
+        position (min (-> most-important inc)
+                      (-> (count headers) dec))]
+    (nth headers (max position 0))))
+
+(defn get-header
+  [type ns]
+  (str "\\" type "{" ns "}"))
+
+;; ## Markdown to LaTeX conversion
+
+(def mark (string/join (repeat 3 "marginalialatex")))
+
+(def separator-marker (str "\n\n" mark "\n\n"))
+
+(def separator-marker-re (re-pattern (str "\n?\n?" mark "\n?\n?")))
+
+(defn invoke-pandoc
+  "We use pandoc to convert `markdown-input` to LaTeX."
+  [markdown-input]
+  (let [{:keys [exit out err]} (shell/sh "pandoc"
+                                         "-f" "markdown" "-t" "latex"
+                                         :in markdown-input)]
+    (when (not= exit 0)
+      (throw
+       (RuntimeException. (str "Error code: " exit ", calling pandoc.\n" err))))
+    out))
+
+(defn md->latex
+  "It applies pandoc in a string joined by `separator-marker`. The
+  output is split using a regex created from `separator-marker`"
+  [docs]
+  (let [joined (string/join separator-marker docs)
+        output (invoke-pandoc joined)
+        result (string/split output separator-marker-re (count docs))]
+    (assert (= (count docs) (count result))
+            "The converted docs must have the same number")
+    result))
+
+(defn to-latex
+  [groups]
+  (map #(assoc-in %1 [:doc] %2) groups (md->latex (map :doc groups))))
+
+;; ## Code and doc extraction
+
+(defn docs-from-group [group]
+  (or (:docstring group)
+      (and (= (:type group) :comment) (:raw group))
+      ""))
+
+(defn code-from-group [group]
+  (if (= (:type group) :code)
+    (:raw group)
+    ""))
+
+(defn extract-code-and-doc
+  "From the original data in each group we extract the code and the
+  documentation."
+  [groups]
+  (map (fn [group] {:code (code-from-group group) :doc (docs-from-group group)})
+       groups))
+
+;; ## Uberdoc generation
+
+(def convert-groups (comp to-latex extract-code-and-doc))
+
+(defn as-data-for-template
+  "
+   1. Converts each `groups` data to the format expected by the template.
+   1. Calculates the correct header level for namespaces."
+  [project-metadata docs]
+  (let [namespaces (map #(update-in % [:groups] convert-groups) docs)
+        section-type (get-section-type (mapcat :groups namespaces))
+        namespaces (map #(assoc % :ns-header (get-header section-type (:ns %)))
+                        namespaces)]
+    {:namespaces namespaces :project project-metadata}))
+
+(defn uberdoc-latex
+  "It uses mustache to generate the LaTeX contents using a
+  template. Before it needs to convert the data to the format expected
+  for the template"
+  [project-metadata docs]
+    (mustache/render template (as-data-for-template project-metadata docs)))
\ No newline at end of file