fix: broaden cv parsing for french labels

Recognize French section labels, allow clear title connector company lines across French and English forms, and keep prose fragments out of experience parsing.

src/job_research/profile/cv_extractor.py

@@ -6,28 +6,34 @@ from pathlib import Path
 from pypdf import PdfReader
 
 
-EXPERIENCE_TITLE_ENDINGS = {
-    "administrator",
-    "analyst",
-    "architect",
-    "consultant",
-    "développeur",
-    "developpeur",
-    "data engineer",
-    "data scientist",
-    "developer",
-    "director",
-    "ingénieur",
-    "ingenieur",
-    "engineer",
-    "lead",
-    "manager",
-    "researcher",
-    "scientist",
-    "specialist",
-}
+EXPERIENCE_LINE_CONNECTORS = (" at ", " chez ", " au ", " à ")
 
-EXPERIENCE_LINE_CONNECTORS = (" at ", " chez ", " au ")
+EXPERIENCE_TITLE_STOPWORDS = {
+    "a",
+    "an",
+    "and",
+    "as",
+    "at",
+    "after",
+    "before",
+    "by",
+    "for",
+    "from",
+    "in",
+    "into",
+    "of",
+    "on",
+    "or",
+    "the",
+    "to",
+    "with",
+    "within",
+    "without",
+    "while",
+    "during",
+    "since",
+    "because",
+}
 
 EXPERIENCE_PROSE_MARKERS = {
     "after",
@@ -55,6 +61,8 @@ EXPERIENCE_PROSE_MARKERS = {
     "until",
     "while",
     "working",
+    "work",
+    "worked",
 }
 
 EDUCATION_TITLE_KEYWORDS = {
@@ -125,7 +133,7 @@ def extract_cv_signals(text: str) -> dict[str, object]:
     for line in non_empty_lines[1:]:
         lowered = line.lower()
 
-        if lowered.startswith("education:"):
+        if lowered.startswith(("education:", "formation:")):
             in_education_section = True
             pending_education_credential = None
             remainder = line.split(":", 1)[1].strip()
@@ -164,10 +172,10 @@ def extract_cv_signals(text: str) -> dict[str, object]:
         if lowered.startswith("location:"):
             location = line.split(":", 1)[1].strip() or None
             continue
-        if lowered.startswith("languages:"):
+        if lowered.startswith(("languages:", "langues:")):
             languages = _parse_csv_field(line)
             continue
-        if lowered.startswith("skills:"):
+        if lowered.startswith(("skills:", "compétences:")):
             skills = _parse_csv_field(line)
             continue
         education_entry = _parse_education_entry(line)
@@ -247,8 +255,7 @@ def _parse_experience_entry(line: str) -> dict[str, str] | None:
         if not title or not company:
             continue
 
-        title_words = title.split()
-        if title_words[-1].lower() not in EXPERIENCE_TITLE_ENDINGS:
+        if not _looks_like_experience_title(title):
             continue
 
         if _looks_like_prose_company(company):
@@ -266,6 +273,23 @@ def _looks_like_experience_line(line: str) -> bool:
     return _parse_experience_entry(line) is not None
 
 
+def _looks_like_experience_title(title: str) -> bool:
+    title_words = [word.strip(".,;:!?()[]{}") for word in title.split()]
+    if not title_words:
+        return False
+
+    first_word = title_words[0]
+    if len(first_word) < 2 or not first_word[0].isupper():
+        return False
+
+    if first_word.lower() in EXPERIENCE_TITLE_STOPWORDS:
+        return False
+
+    return not any(
+        word.lower() in EXPERIENCE_PROSE_MARKERS for word in title_words if word
+    )
+
+
 def _looks_like_prose_company(company: str) -> bool:
     company_words = [word.strip(".,;:!?()[]{}") for word in company.split()]
     return any(

tests/profile/test_cv_extractor.py

@@ -107,6 +107,26 @@ def test_extract_cv_signals_normalizes_en_dash_bullet_prefixed_fields_and_experi
     ]
 
 
+def test_extract_cv_signals_recognizes_french_field_labels() -> None:
+    text = dedent(
+        """
+        Tonio
+        Formation:
+        M.Sc. in Data Engineering at EPITA
+        Langues: French, English
+        Compétences: Python, SQL
+        """
+    ).strip()
+
+    extracted = extract_cv_signals(text)
+
+    assert extracted["languages"] == ["French", "English"]
+    assert extracted["skills"] == ["Python", "SQL"]
+    assert extracted["education_entries"] == [
+        {"credential": "M.Sc. in Data Engineering", "institution": "EPITA"}
+    ]
+
+
 def test_extract_cv_signals_parses_french_experience_connectors() -> None:
     text = dedent(
         """
@@ -125,6 +145,30 @@ def test_extract_cv_signals_parses_french_experience_connectors() -> None:
     ]
 
 
+def test_extract_cv_signals_parses_clear_titles_with_french_and_english_connectors() -> None:
+    text = dedent(
+        """
+        Tonio
+        Location: France
+        Ingénieur à Thales
+        Ingénieur Data chez BNP Paribas
+        Consultant BI chez Accenture
+        Head of Data at Qonto
+        Product Owner at Qonto
+        """
+    ).strip()
+
+    extracted = extract_cv_signals(text)
+
+    assert extracted["experience_entries"] == [
+        {"title": "Ingénieur", "company": "Thales"},
+        {"title": "Ingénieur Data", "company": "BNP Paribas"},
+        {"title": "Consultant BI", "company": "Accenture"},
+        {"title": "Head of Data", "company": "Qonto"},
+        {"title": "Product Owner", "company": "Qonto"},
+    ]
+
+
 def test_extract_cv_signals_ignores_french_prose_continuations() -> None:
     text = dedent(
         """