diff --git a/lilac/signals/default_signals.py b/lilac/signals/default_signals.py
index 8464819ce..a8d5834f8 100644
--- a/lilac/signals/default_signals.py
+++ b/lilac/signals/default_signals.py
@@ -10,6 +10,7 @@
 from .concept_labels import ConceptLabelsSignal
 from .concept_scorer import ConceptSignal
 from .lang_detection import LangDetectionSignal
+from .markdown_code_block import MarkdownCodeBlockSignal
 from .near_dup import NearDuplicateSignal
 from .ner import SpacyNER
 from .pii import PIISignal
@@ -29,6 +30,7 @@ def register_default_signals() -> None:
   register_signal(NearDuplicateSignal)
   register_signal(LangDetectionSignal)
   register_signal(ClusterHDBScan)
+  register_signal(MarkdownCodeBlockSignal)
 
   # Embeddings.
   register_signal(Cohere)
diff --git a/lilac/signals/markdown_code_block.py b/lilac/signals/markdown_code_block.py
new file mode 100644
index 000000000..d0ad775ce
--- /dev/null
+++ b/lilac/signals/markdown_code_block.py
@@ -0,0 +1,47 @@
+"""Finds markdown code blocks.
+
+NOTE: It would be great to use guesslang to detect the language automatically, however
+there is a dependency conflict with typing extensions.
+"""
+import re
+from typing import ClassVar, Iterable, Iterator, Optional, cast
+
+from typing_extensions import override
+
+from ..schema import Field, Item, RichData, field, span
+from ..signal import TextSignal
+
+MARKDOWN_RE = '```([^\n ]*?)\n(.*?)\n```'
+
+
+class MarkdownCodeBlockSignal(TextSignal):
+  """Finds markdown blocks in text. Emits the language of the block with the span."""
+
+  name: ClassVar[str] = 'markdown_code_block'
+  display_name: ClassVar[str] = 'Markdown Code Block Detection'
+
+  @override
+  def fields(self) -> Field:
+    return field(
+      fields=[
+        field(
+          dtype='string_span',
+          fields={'language': 'string'},
+        )
+      ]
+    )
+
+  @override
+  def compute(self, data: Iterable[RichData]) -> Iterator[Optional[Item]]:
+    markdown_re = re.compile(MARKDOWN_RE, re.MULTILINE | re.DOTALL)
+    for doc in data:
+      text = cast(str, doc)
+      # Get the spans
+      markdown_re_spans = markdown_re.finditer(text)
+      languages = markdown_re.findall(text)
+
+      spans: list[Item] = []
+      for re_span, (language, _) in zip(markdown_re_spans, languages):
+        spans.append(span(re_span.start(), re_span.end(), {'language': language}))
+
+      yield spans
diff --git a/lilac/signals/markdown_code_block_test.py b/lilac/signals/markdown_code_block_test.py
new file mode 100644
index 000000000..3c9a6bd88
--- /dev/null
+++ b/lilac/signals/markdown_code_block_test.py
@@ -0,0 +1,82 @@
+"""Test the Markdown Extractor signal."""
+
+from ..schema import field
+from ..splitters.text_splitter_test_utils import text_to_expected_spans
+from .markdown_code_block import MarkdownCodeBlockSignal
+
+
+def test_markdown_code_block_fields() -> None:
+  signal = MarkdownCodeBlockSignal()
+  signal.setup()
+  assert signal.fields() == field(fields=[field('string_span', fields={'language': 'string'})])
+
+
+def test_markdown_code_block() -> None:
+  signal = MarkdownCodeBlockSignal()
+  signal.setup()
+
+  text = """
+I am trying to add an extra field to my model form in Django.
+
+```python
+class MyForm(forms.ModelForm):
+    extra_field = forms.CharField()
+    class Meta:
+        model = MyModel
+        widgets = {
+            'extra_field': forms.Textarea(attrs={'placeholder': u'Bla bla'}),
+        }
+```
+However, it appears that the widget definition for 'extra_field' in the Meta class is ignored.
+
+```py
+class MyForm(forms.ModelForm):
+    extra_field = forms.CharField(widget=forms.Textarea())
+    class Meta:
+        model = MyModel
+```
+
+Could you explain why my first approach does not work and what I am doing wrong?
+
+Here is the console output:
+```
+fake output
+```
+
+"""
+  markdown_blocks = list(signal.compute([text]))
+
+  expected_spans = text_to_expected_spans(
+    text,
+    [
+      (
+        """```python
+class MyForm(forms.ModelForm):
+    extra_field = forms.CharField()
+    class Meta:
+        model = MyModel
+        widgets = {
+            'extra_field': forms.Textarea(attrs={'placeholder': u'Bla bla'}),
+        }
+```""",
+        {'language': 'python'},
+      ),
+      (
+        """```py
+class MyForm(forms.ModelForm):
+    extra_field = forms.CharField(widget=forms.Textarea())
+    class Meta:
+        model = MyModel
+```""",
+        {'language': 'py'},
+      ),
+      (
+        """```
+fake output
+```""",
+        {'language': ''},
+      ),
+    ],
+  )
+
+  assert markdown_blocks == [expected_spans]
diff --git a/notebooks/CurateCodingDataset.ipynb b/notebooks/CurateCodingDataset.ipynb
index fedfc006d..3718b6f1c 100644
--- a/notebooks/CurateCodingDataset.ipynb
+++ b/notebooks/CurateCodingDataset.ipynb
@@ -193,7 +193,7 @@
     "import lilac as ll\n",
     "from pprint import pprint\n",
     "\n",
-    "code_block_re = re.compile('```(py|python)\\n(.*)?\\n```', re.MULTILINE | re.DOTALL)\n",
+    "code_block_re = re.compile('```(py|python)\\n(.*?)\\n```', re.MULTILINE | re.DOTALL)\n",
     "\n",
     "\n",
     "# Format the code blocks of the \"answer\" column using the `ruff`` formatter.\n",
diff --git a/web/blueprint/src/lib/components/datasetView/ItemMediaTextContent.svelte b/web/blueprint/src/lib/components/datasetView/ItemMediaTextContent.svelte
index 4e49be103..57b6c0913 100644
--- a/web/blueprint/src/lib/components/datasetView/ItemMediaTextContent.svelte
+++ b/web/blueprint/src/lib/components/datasetView/ItemMediaTextContent.svelte
@@ -511,7 +511,7 @@
               }
             }
           });
-        } else if (renderSpan.isLeafSpan) {
+        } else if (renderSpan.isMetadata) {
           spanDecorations.push({
             range,
             options: {
@@ -678,7 +678,7 @@
     @apply border-r-8 border-orange-700 opacity-20;
   }
   :global(.leaf-text) {
-    @apply font-extrabold text-violet-500 underline;
+    @apply text-violet-500;
   }
 
   /** Deep-linked selection */
diff --git a/web/blueprint/src/lib/components/datasetView/spanHighlight.ts b/web/blueprint/src/lib/components/datasetView/spanHighlight.ts
index 252ace943..9c2778f56 100644
--- a/web/blueprint/src/lib/components/datasetView/spanHighlight.ts
+++ b/web/blueprint/src/lib/components/datasetView/spanHighlight.ts
@@ -57,6 +57,7 @@ export interface MonacoRenderSpan {
   isKeywordSearch: boolean;
   isConceptSearch: boolean;
   isSemanticSearch: boolean;
+  isMetadata: boolean;
   isLeafSpan: boolean;
   hasNonNumericMetadata: boolean;
 
@@ -103,6 +104,7 @@ export function getMonacoRenderSpans(
         const isConceptSearch = valueInfo.type === 'concept_score';
         const isSemanticSearch = valueInfo.type === 'semantic_similarity';
         const hasNonNumericMetadata = valueInfo.type === 'metadata' && !isNumeric(valueInfo.dtype);
+        const isMetadata = valueInfo.type === 'metadata';
         const isLeafSpan = valueInfo.type === 'leaf_span';
 
         let isHighlighted = false;
@@ -124,6 +126,7 @@ export function getMonacoRenderSpans(
           isKeywordSearch,
           isConceptSearch,
           isSemanticSearch,
+          isMetadata,
           isLeafSpan,
           hasNonNumericMetadata,
           namedValue,