commitred with correct city name

backend/employment_agent/decorators.py

@@ -103,17 +103,14 @@ def _clean_phone(value,resume_text):
 
 
 def _clean_city(value,resume_text):
-    """Optional residence city. Sentinel / invented → None. Never rejects the CV.
+    """Optional residence city. Sentinel → None. Never rejects the CV.
 
-    The prompt forbids work-experience cities; this only drops a value that is
-    absent from the resume text or is the explicit empty sentinel.
+    Proper city names (Karachi, not Karachi(Malir)) come from the OpenAI parse
+    in run_employment_agent. This clamp does not rewrite place names.
     """
     text=(value or "").strip()
     if not text or text.lower() in (NO_CITY.lower(),"none","null","n/a","-"):
         return None
-    haystack=(resume_text or "").lower()
-    if haystack and text.lower() not in haystack:
-        return None
     return text
 
 

backend/employment_agent/execute_agent.py

@@ -6,15 +6,67 @@ Called from inbox.tasks.match_inbox_message; no HTTP surface.
 
 from __future__ import annotations
 
+import json
 import logging
 
 from employment_agent.decorators import parse_employment_response
-from employment_agent.prompt import CURRENT_TITLE,EDUCATION,NO_COMPANY,prompt,user_prompt
+from employment_agent.prompt import CURRENT_TITLE,EDUCATION,NO_CITY,NO_COMPANY,city_list_prompt,prompt,user_prompt
 from llm_setup import llm_call
 
 logger=logging.getLogger("employment_agent")
 
 
+def parse_normalized_cities(data,fallback=None):
+    """Keep unique proper city names from the list-normalizer JSON."""
+    rows=None
+    if isinstance(data,dict):
+        rows=data.get("cities")
+    if not isinstance(rows,list):
+        return list(fallback or [])
+    out=[]
+    seen=set()
+    for item in rows:
+        if not isinstance(item,str):
+            continue
+        text=item.strip()
+        if not text or text.lower() in (NO_CITY.lower(),"none","null","n/a","-"):
+            continue
+        key=text.lower()
+        if key in seen:
+            continue
+        seen.add(key)
+        out.append(text)
+    out.sort(key=str.lower)
+    return out or list(fallback or [])
+
+
+async def normalize_cities(values):
+    """OpenAI: messy stored places → the same proper city names the CV agent writes."""
+    places=[]
+    seen=set()
+    for raw in values or []:
+        text=(raw or "").strip()
+        if not text:
+            continue
+        key=text.lower()
+        if key in seen:
+            continue
+        seen.add(key)
+        places.append(text)
+    if not places:
+        return []
+    try:
+        data=await llm_call(
+            city_list_prompt(),
+            json.dumps({"places":places},ensure_ascii=False),
+            json_mode=True,
+        )
+    except Exception:
+        logger.exception("city list normalize failed")
+        return places
+    return parse_normalized_cities(data,fallback=places)
+
+
 async def run_employment_agent(*,resume_text=""):
     text=(resume_text or "").strip()
     if not text:

backend/employment_agent/prompt.py

@@ -14,6 +14,14 @@ NO_LINKEDIN="no linkedin url mentioned"
 NO_PHONE="no phone number mentioned"
 NO_CITY="no city mentioned"
 
+CITY_POLICY=f"""- Return ONE proper city name only — the city, not an area, town, sector, housing society, cantonment, district, or parenthetical locality.
+- Correct values look like "Karachi", "Lahore", "Islamabad", "Rawalpindi", "Peshawar". Not "Karachi(Malir)", not "Wah Cantt", not "Gulberg Lahore".
+- If the text names a neighborhood or area of a city, return that city: "Karachi (Malir)" / "Karachi(Malir)" / "DHA Karachi" → "Karachi". "Gulberg, Lahore" → "Lahore". "F-10 Islamabad" → "Islamabad".
+- Drop "Cantt" / "Cantonment": "Lahore Cantt" → "Lahore", "Rawalpindi Cantt" → "Rawalpindi", "Wah Cantt" → "Wah".
+- Never concatenate two places. If the string is messy (for example "Karachi(Malir) Wah Cantt"), return the single residence city, not both strings glued together.
+- Do not return province, country, street, house number, or text inside parentheses.
+- Drop junk tokens such as KA, KAR, KARA, empty values, and unintelligible strings."""
+
 
 def prompt():
     return f"""You are an HR-ATS recruiting assistant.
@@ -57,11 +65,10 @@ linkedin_url (its own key — extract this separately from the other fields):
 - Never guess a slug or construct linkedin.com/in/<name> from the candidate's name. The stored value will be null when this sentinel is returned.
 
 city (its own key — OPTIONAL. A missing city must not fail the candidate):
-- Return the city of residence only, city name alone (for example "Karachi", "Lahore", "Islamabad"), using the resume's own spelling.
+{CITY_POLICY}
 - Extract city ONLY from the candidate's contact / location / address header (the block with name, phone, email, LinkedIn, "Address", "Location", "based in", "currently living in").
 - Do NOT extract city from Work Experience. A job that lists Karachi, UAE, USA, or any other city is the employer's location, not proof the candidate lives there.
 - If the contact/location section does not name a city, return exactly: {NO_CITY}. Leave it blank rather than guessing from jobs, education, or nationality.
-- The city string you return MUST appear verbatim (or as a clear substring) in that contact/location section of the resume text.
 
 phone (its own key — extract this separately; copy EVERY digit):
 - Return the candidate's own mobile / phone exactly as written, including country code when present.
@@ -118,6 +125,14 @@ Example 9 — contact/location city is residence:
 Resume: "Ali Khan | Location: Lahore | 0321-5551234\\nExperience: Acme, Karachi, Engineer"
 JSON city must be "Lahore". Not "Karachi".
 
+Example 10 — neighborhood / cantonment is not the city:
+Resume: "Ali Khan | Karachi(Malir) | 0321-5551234"
+JSON city must be "Karachi". Not "Karachi(Malir)" and not "Malir".
+
+Example 11 — do not glue two place fragments:
+Resume: "Address: Karachi(Malir) Wah Cantt"
+JSON city must be "Karachi" (one city). Not "Karachi(Malir) Wah Cantt" and not "Wah Cantt".
+
 Respond with JSON only:
 {{
   "current_employment": "Company Name",
@@ -135,3 +150,16 @@ If the contact/location section has no city, city must be "{NO_CITY}" — still
 
 def user_prompt(resume_text:str) -> str:
     return json.dumps({"resume_text":resume_text or ""},ensure_ascii=False)
+
+
+def city_list_prompt():
+    """Map messy stored place strings to the same proper city names the CV agent writes."""
+    return f"""You map messy residence strings to proper city names for an Inbox City filter.
+
+{CITY_POLICY}
+
+Input is JSON: {{"places": ["Karachi(Malir)", "Wah Cantt", "Lahore"]}}
+Respond with JSON only:
+{{"cities": ["Karachi", "Wah", "Lahore"]}}
+Unique proper city names only. Do not copy raw neighborhood or cantonment strings into cities.
+"""

backend/g_sheet/models.py

@@ -126,6 +126,18 @@ class FormData(SQLModel, table=True):
             else_=False,
         )
 
+    @staticmethod
+    def _cities_match(column, cities):
+        """Agent city name vs stored raw text: Karachi matches Karachi(Malir)."""
+        clauses = []
+        for city in cities or []:
+            text = (city or "").strip()
+            if not text:
+                continue
+            safe = text.replace("\\", "\\\\").replace("%", "\\%").replace("_", "\\_")
+            clauses.append(column.ilike(f"%{safe}%", escape="\\"))
+        return or_(*clauses) if clauses else None
+
     @classmethod
     def _filters(
         cls, *, sheet=None, search=None, processing_state=None, is_duplicate=None,
@@ -160,8 +172,9 @@ class FormData(SQLModel, table=True):
             )
         cities = [c.strip() for c in (city or []) if (c or "").strip()]
         if cities:
-            city_col = func.lower(func.coalesce(cls.city, cls.residing_city))
-            filters.append(city_col.in_([c.lower() for c in cities]))
+            clause = cls._cities_match(func.coalesce(cls.city, cls.residing_city), cities)
+            if clause is not None:
+                filters.append(clause)
         if source:
             text = source.strip()
             lowered = text.lower()

backend/inbox/models.py

@@ -1022,6 +1022,18 @@ class Inbox_Messages(SQLModel, table=True):
             else_=False,
         )
 
+    @staticmethod
+    def _cities_match(column, cities):
+        """Agent city name vs stored raw text: Karachi matches Karachi(Malir)."""
+        clauses = []
+        for city in cities or []:
+            text = (city or "").strip()
+            if not text:
+                continue
+            safe = text.replace("\\", "\\\\").replace("%", "\\%").replace("_", "\\_")
+            clauses.append(column.ilike(f"%{safe}%", escape="\\"))
+        return or_(*clauses) if clauses else None
+
     @classmethod
     def _apply_filters(
         cls, statement, search: str | None=None, isread: bool=True,
@@ -1061,7 +1073,9 @@ class Inbox_Messages(SQLModel, table=True):
             statement = statement.where(cls.processing_state == processing_state)
         cities = [c.strip() for c in (city or []) if (c or "").strip()]
         if cities:
-            statement = statement.where(func.lower(cls.city).in_([c.lower() for c in cities]))
+            clause = cls._cities_match(cls.city, cities)
+            if clause is not None:
+                statement = statement.where(clause)
         if source:
             text = source.strip()
             lowered = text.lower()

backend/inbox/views.py

@@ -513,11 +513,13 @@ class Email:
             return await Inbox_Messages.count_inbox_messages(self.session,search,assigned=assigned,is_duplicate=is_duplicate,no_suggestions=no_suggestions,processing_state=processing_state,city=city,source=source)
 
     async def list_cities(self):
-        """Distinct cities from inbox_messages and form_data, merged in Python."""
+        """Proper city names for the Inbox filter — same OpenAI mapping as CV parse."""
+        from employment_agent.execute_agent import normalize_cities
         from g_sheet.models import FormData
         inbox=await Inbox_Messages.distinct_cities(self.session)
         forms=await FormData.distinct_cities(self.session)
-        return Reapplied(session=self.session).merge_cities(inbox,forms)
+        merged=Reapplied(session=self.session).merge_cities(inbox,forms)
+        return await normalize_cities(merged)
 
     async def list_sources(self):
         """Source / platform labels: seeded channels, Google Sheet, form sources."""

backend/tests/test_employment_agent.py

@@ -89,3 +89,27 @@ def test_adds_scheme_and_rejects_company_page():
         "",
     )
     assert company_page["linkedin_url"] is None
+
+
+def test_city_prompt_asks_openai_for_a_proper_city_name():
+    from employment_agent.prompt import city_list_prompt, prompt
+    text = prompt()
+    assert "Karachi(Malir)" in text
+    assert 'JSON city must be "Karachi"' in text
+    assert "Return ONE proper city name only" in text
+    listed = city_list_prompt()
+    assert "Karachi(Malir)" in listed
+    assert '"cities"' in listed
+
+
+def test_parse_normalized_cities_keeps_agent_names():
+    from employment_agent.execute_agent import parse_normalized_cities
+    assert parse_normalized_cities(
+        {"cities": ["Karachi", "Karachi", "Lahore", "no city mentioned"]},
+        fallback=["Karachi(Malir)"],
+    ) == ["Karachi", "Lahore"]
+
+
+def test_parse_normalized_cities_falls_back_when_the_model_shape_is_wrong():
+    from employment_agent.execute_agent import parse_normalized_cities
+    assert parse_normalized_cities({"oops": True}, fallback=["Karachi(Malir)"]) == ["Karachi(Malir)"]

backend/tests/test_employment_extraction_clamps.py

@@ -163,3 +163,13 @@ def test_existing_sentinels_still_normalize():
     assert fields["current_employment"] == NO_COMPANY
     assert fields["education"] == EDUCATION
     assert fields["skills"] == ["Python"]
+
+
+def test_city_from_the_model_is_kept_as_returned():
+    resume = "Ali Khan | Karachi(Malir) | 0321-5551234"
+    fields = parse_employment_response({"city": "Karachi"}, resume)
+    assert fields["city"] == "Karachi"
+
+
+def test_city_sentinel_is_dropped():
+    assert parse({"city": "no city mentioned"})["city"] is None

backend/tests/test_form_data_filters.py

@@ -96,6 +96,13 @@ class TestListAndBadgesAgree:
             assert name in params, f"count_processing should narrow on {name}"
 
 
+class TestCity:
+    def test_agent_city_matches_raw_stored_text(self):
+        out = sql(*FormData._filters(city=["Karachi"]))
+        assert "ilike" in out
+        assert "karachi" in out
+
+
 class TestPlumbing:
     @pytest.mark.parametrize(
         "func", [FormData.fetch_form_data, FormData.count_form_data],

frontend/src/screens/Inbox.jsx

@@ -977,7 +977,7 @@ function BulkReadBar({ rows, selection, onSetRead, onSetAllRead, busy, canEdit,
  * than a refactor of theirs.)
  */
 /** Junk location tokens — hide from the dropdown, not from the list query. */
-const HIDDEN_LOCATION = /^(KA|KAR|KARA|WAH)$/i
+const HIDDEN_LOCATION = /^(KA|KAR|KARA)$/i
 
 function InboxFilters({ filters, cities, sources, showLinkFilters, active, open, onToggle, onChange, onClear }) {
   const locations = (cities || []).filter((name) => !HIDDEN_LOCATION.test(String(name).trim()))