fix: relax cv title parsing

Allow lowercase, mixed-case, and numeric experience titles while rejecting label-like lines without colons via a small blocked-label set.

src/job_research/profile/cv_extractor.py

@@ -35,6 +35,20 @@ EXPERIENCE_TITLE_STOPWORDS = {
     "because",
 }
 
+EXPERIENCE_TITLE_LABELS = {
+    "education",
+    "experience",
+    "formation",
+    "languages",
+    "langues",
+    "location",
+    "profile",
+    "skills",
+    "compétences",
+    "competences",
+    "summary",
+}
+
 EXPERIENCE_PROSE_MARKERS = {
     "after",
     "before",
@@ -279,14 +293,18 @@ def _looks_like_experience_title(title: str) -> bool:
         return False
 
     first_word = title_words[0]
-    if len(first_word) < 2 or not first_word[0].isupper():
+    if len(first_word) < 2:
         return False
 
-    if first_word.lower() in EXPERIENCE_TITLE_STOPWORDS:
+    normalized_title = " ".join(title_words).casefold()
+    if normalized_title in EXPERIENCE_TITLE_LABELS:
+        return False
+
+    if first_word.casefold() in EXPERIENCE_TITLE_STOPWORDS:
         return False
 
     return not any(
-        word.lower() in EXPERIENCE_PROSE_MARKERS for word in title_words if word
+        word.casefold() in EXPERIENCE_PROSE_MARKERS for word in title_words if word
     )
 
 

tests/profile/test_cv_extractor.py

@@ -169,6 +169,26 @@ def test_extract_cv_signals_parses_clear_titles_with_french_and_english_connecto
     ]
 
 
+def test_extract_cv_signals_accepts_lowercase_mixed_case_and_numeric_titles() -> None:
+    text = dedent(
+        """
+        Tonio
+        Location: France
+        data engineer at Company A
+        iOS Engineer at Company A
+        3D Artist at Studio
+        """
+    ).strip()
+
+    extracted = extract_cv_signals(text)
+
+    assert extracted["experience_entries"] == [
+        {"title": "data engineer", "company": "Company A"},
+        {"title": "iOS Engineer", "company": "Company A"},
+        {"title": "3D Artist", "company": "Studio"},
+    ]
+
+
 def test_extract_cv_signals_ignores_french_prose_continuations() -> None:
     text = dedent(
         """
@@ -183,6 +203,22 @@ def test_extract_cv_signals_ignores_french_prose_continuations() -> None:
     assert extracted["experience_entries"] == []
 
 
+def test_extract_cv_signals_rejects_label_like_lines_without_colons() -> None:
+    text = dedent(
+        """
+        Tonio
+        Location at Paris
+        Summary at a glance
+        Profile at LinkedIn
+        Education at EPITA
+        """
+    ).strip()
+
+    extracted = extract_cv_signals(text)
+
+    assert extracted["experience_entries"] == []
+
+
 def test_extract_cv_signals_ignores_in_paris_prose_tail() -> None:
     text = dedent(
         """