fix: tighten cv extraction heuristics

src/job_research/profile/cv_extractor.py

@@ -5,9 +5,32 @@ from pathlib import Path
 from pypdf import PdfReader
 
 
+EXPERIENCE_TITLE_ENDINGS = {
+    "administrator",
+    "analyst",
+    "architect",
+    "consultant",
+    "data engineer",
+    "data scientist",
+    "developer",
+    "director",
+    "engineer",
+    "lead",
+    "manager",
+    "researcher",
+    "scientist",
+    "specialist",
+}
+
+
 def extract_pdf_text(path: Path) -> str:
     reader = PdfReader(str(path))
-    return "\n".join(page.extract_text() or "" for page in reader.pages)
+    page_texts = [
+        text.strip()
+        for text in (page.extract_text() or "" for page in reader.pages)
+        if text.strip()
+    ]
+    return "\n".join(page_texts)
 
 
 def extract_cv_signals(text: str) -> dict[str, object]:
@@ -31,7 +54,7 @@ def extract_cv_signals(text: str) -> dict[str, object]:
         if lowered.startswith("skills:"):
             skills = _parse_csv_field(line)
             continue
-        if " at " in line:
+        if _looks_like_experience_line(line):
             title, company = line.split(" at ", 1)
             experience_entries.append(
                 {
@@ -53,3 +76,21 @@ def extract_cv_signals(text: str) -> dict[str, object]:
 def _parse_csv_field(line: str) -> list[str]:
     _, value = line.split(":", 1)
     return [item.strip() for item in value.split(",") if item.strip()]
+
+
+def _looks_like_experience_line(line: str) -> bool:
+    if line.count(" at ") != 1:
+        return False
+
+    title, company = (part.strip() for part in line.split(" at ", 1))
+    if not title or not company:
+        return False
+
+    title_words = title.split()
+    if len(title_words) < 2:
+        return False
+
+    if title_words[-1].lower() not in EXPERIENCE_TITLE_ENDINGS:
+        return False
+
+    return any(character.isupper() for character in company)

tests/profile/test_cv_extractor.py

@@ -1,6 +1,8 @@
 from textwrap import dedent
+from pathlib import Path
 
 from job_research.profile.cv_extractor import extract_cv_signals
+from job_research.profile.cv_extractor import extract_pdf_text
 
 
 def test_extract_cv_signals_reads_basic_fields_from_text() -> None:
@@ -23,3 +25,49 @@ def test_extract_cv_signals_reads_basic_fields_from_text() -> None:
     assert extracted["skills"] == ["Python", "SQL", "Terraform", "GCP", "BigQuery"]
     assert extracted["experience_entries"][0]["title"] == "Data Engineer"
     assert len(extracted["experience_entries"]) == 2
+
+
+def test_extract_cv_signals_ignores_prose_that_mentions_at() -> None:
+    text = dedent(
+        """
+        Tonio
+        Location: France
+        Worked at a startup before joining the team.
+        """
+    ).strip()
+
+    extracted = extract_cv_signals(text)
+
+    assert extracted["experience_entries"] == []
+
+
+def test_extract_pdf_text_skips_blank_pages(monkeypatch) -> None:
+    class FakePage:
+        def __init__(self, text: str | None) -> None:
+            self._text = text
+
+        def extract_text(self) -> str | None:
+            return self._text
+
+    class FakePdfReader:
+        def __init__(self, path: str) -> None:
+            self.path = path
+            self.pages = [
+                FakePage(None),
+                FakePage("Tonio"),
+                FakePage(""),
+                FakePage("Data Engineer"),
+            ]
+
+    seen_paths: list[str] = []
+
+    def fake_pdf_reader(path: str) -> FakePdfReader:
+        seen_paths.append(path)
+        return FakePdfReader(path)
+
+    monkeypatch.setattr("job_research.profile.cv_extractor.PdfReader", fake_pdf_reader)
+
+    extracted = extract_pdf_text(Path("cv.pdf"))
+
+    assert extracted == "Tonio\nData Engineer"
+    assert seen_paths == ["cv.pdf"]