Update app.py

app.py

@@ -58,6 +58,7 @@ intent_tokenizer = intent_model = label_encoder = id2intent = None
 ner_pipeline     = label2id = id2label = None
 semantic_model   = corpus_df = corpus_embeddings = places_df = None
 SESSIONS: dict   = {}
+MENU_CACHE: dict  = {}
 STARTUP_COMPLETE: bool = False
 _SESSION_TTL_SEC = 3600   # session بتتمسح بعد ساعة من غير استخدام
 
@@ -1582,7 +1583,7 @@ def chat(text: str, session, user_lat=None, user_lon=None):
     _LOC_REQ = ["موقعي","لوكيشن بتاعي","استخدم موقعي","تعرف تجيب لوكيشن",
                 "موقعك","location بتاعي","gps"]
     if any(w in t_norm for w in _LOC_REQ):
-        if user_lat and user_lon:
+        if user_lat is not None and user_lon is not None:
             reply = "تمام، هستخدم موقعك عشان أجيب الأقرب ليك. 📍"
         else:
             reply = "محتاج تسمحلي باللوكيشن من التطبيق الأول. 📍"
@@ -1759,7 +1760,7 @@ def chat(text: str, session, user_lat=None, user_lon=None):
 
     # ── proximity query: لو مفيش lat/lon ────────────────────────
     t_low = norm(text)
-    if _is_proximity_query(text) and not (user_lat and user_lon):
+    if _is_proximity_query(text) and not (user_lat is not None and user_lon is not None):
         reply = ("📍 عشان أحدد الأقرب محتاج تسمحلي باللوكيشن.\n"
                  "أو قولي المنطقة اللي أنت فيها.")
         result.update(reply=reply, intent="missing_info",
@@ -1956,17 +1957,69 @@ _ITEM_SEARCH_PATTERNS = [
 ]
 
 def _looks_like_item_search(text: str) -> bool:
-    """Detect item/menu-item search even if the item word is not in our keywords."""
+    """
+    Detect item/menu-item search without stealing normal category requests.
+
+    Important fixes:
+    - "عايزة سوبرماركت" is nearest_supermarket, not search_by_item.
+    - "عايزة احسن مطعم" is recommendation/search, not item_query=احسن.
+    - "عايزة برجر" / "مكان عنده اسموزي خوخ" stays search_by_item.
+    """
     t = norm(text)
-    # menu display is different from item search
-    if any(kw in t for kw in _MENU_TRIGGERS) and not any(p in t for p in ["عنده", "عندها", "بيعمل", "بتعمل", "بيقدم", "بتقدم", "فيه", "فيها"]):
+
+    # menu display is different from item search unless the user asks for a place that has an item
+    item_context_words = [
+        "عنده", "عندها", "بيعمل", "بتعمل", "بيقدم", "بتقدم",
+        "فيه", "فيها", "مكان عنده", "مكان عندها", "مطعم عنده", "كافيه عنده",
+        "has", "serves", "serve"
+    ]
+    if any(kw in t for kw in _MENU_TRIGGERS) and not any(norm(p) in t for p in item_context_words):
         return False
-    has_item_pattern = any(norm(p) in t for p in _ITEM_SEARCH_PATTERNS)
-    category_context = infer_category(text) in ("restaurant", "cafe")
-    has_item_keyword = any(norm(kw) in t for kw in _ITEM_TRIGGERS)
+
+    inferred_cat = infer_category(text)
     candidate = extract_item_query(text)
-    # لو فيه صيغة طلب + باقي نص بعد الشيل يبقى غالبًا صنف، حتى لو مش في keywords
-    return bool(has_item_keyword or (has_item_pattern and candidate) or (category_context and has_item_pattern and candidate))
+    cand_clean = clean_text(candidate or "")
+
+    # If the remaining candidate is only a category or a ranking adjective, do NOT treat it as food item.
+    category_words = set()
+    for words in CATEGORY_KEYWORDS.values():
+        category_words.update(clean_text(w) for w in words)
+    category_words.update(clean_text(k) for k in _CAT_MAP.keys())
+
+    ranking_words = {
+        "احسن", "افضل", "الاحسن", "الافضل", "أحسن", "أفضل",
+        "كويس", "كويسة", "حلو", "حلوة", "ترشيح", "رشحلي",
+        "best", "good", "recommended", "recommend"
+    }
+    ranking_words = {clean_text(w) for w in ranking_words}
+
+    cand_tokens = [w for w in cand_clean.split() if w]
+    if cand_tokens and all((w in category_words or w in ranking_words) for w in cand_tokens):
+        return False
+
+    # Pure category requests must go to nearest_* intents, not item search.
+    if inferred_cat in ("supermarket", "pharmacy", "housing"):
+        return False
+    if inferred_cat in ("restaurant", "cafe") and cand_tokens and all(w in category_words or w in ranking_words for w in cand_tokens):
+        return False
+
+    has_item_keyword = any(norm(kw) in t for kw in _ITEM_TRIGGERS)
+    has_item_pattern = any(norm(p) in t for p in _ITEM_SEARCH_PATTERNS)
+    has_strong_item_context = any(norm(p) in t for p in item_context_words)
+
+    # Known item names are enough: "عايزة برجر".
+    if has_item_keyword:
+        return True
+
+    # Unknown item names need strong item context: "مكان عنده فتة" / "كافيه بيعمل موهي��و".
+    if has_strong_item_context and candidate and len(cand_clean) >= 3:
+        return True
+
+    # Avoid broad patterns like "عايزة" stealing normal searches.
+    if has_item_pattern and inferred_cat in ("restaurant", "cafe") and candidate and len(cand_clean) >= 3:
+        return False
+
+    return False
 
 def _detect_new_intent(text: str) -> str | None:
     """
@@ -2329,6 +2382,97 @@ def get_place_menu(place_id: str) -> list[dict]:
         log.error(f"❌ get_menu error: {e}")
         return []
 
+
+
+def _place_id_for_menu(row: dict) -> str | None:
+    """Pick a usable id for /places/{place_id}/menu from mixed DB/API rows."""
+    for k in ("place_id", "id", "placeId"):
+        v = row.get(k) if isinstance(row, dict) else None
+        if v is not None and str(v).strip() and str(v).strip().lower() != "nan":
+            return str(v).strip()
+    return None
+
+
+def _get_menu_cached(place_id: str) -> list[dict]:
+    """Small in-memory cache so item search does not re-download menus every turn."""
+    if not place_id:
+        return []
+    if place_id in MENU_CACHE:
+        return MENU_CACHE[place_id]
+    items = get_place_menu(place_id)
+    MENU_CACHE[place_id] = items
+    return items
+
+
+def _infer_item_kind(item_query: str) -> str:
+    """Return food/drink/unknown without depending on a closed keyword list only."""
+    q = clean_text(item_query)
+    drink_hints = {
+        "قهوه", "قهوة", "كوفي", "لاتيه", "اسبريسو", "نسكافيه", "كابتشينو",
+        "عصير", "مشروب", "مشروبات", "موهيتو", "ميلك", "شيك", "ايس", "آيس",
+        "سموزي", "اسموزي", "سموذي", "smoothie", "juice", "coffee", "latte", "tea", "شاي"
+    }
+    food_hints = {
+        "برجر", "بيتزا", "شاورما", "كريب", "فراخ", "فرخه", "دجاج", "مشويات", "كباب",
+        "كفته", "كفتة", "سمك", "سندوتش", "ساندوتش", "حواوشي", "كشري", "مكرونه", "وجبه",
+        "burger", "pizza", "shawarma", "crepe", "chicken", "grill", "sandwich"
+    }
+    if any(clean_text(w) in q for w in drink_hints):
+        return "drink"
+    if any(clean_text(w) in q for w in food_hints):
+        return "food"
+    return "unknown"
+
+
+def scan_menus_for_item(item_query: str, max_places_to_scan: int = 120) -> list[dict]:
+    """
+    Last-resort but important fallback:
+    Search inside actual menus when /search-by-item returns empty or too strict.
+    This fixes cases where the item exists in menu but is not present in item keywords
+    or the backend search endpoint does not match the spelling.
+    """
+    if places_df is None or places_df.empty:
+        return []
+
+    kind = _infer_item_kind(item_query)
+    df = places_df.copy()
+
+    if "category_clean" in df.columns:
+        if kind == "food":
+            df = df[df["category_clean"].astype(str).apply(lambda x: normalize_category(x) == "restaurant" or "مطعم" in x)]
+        elif kind == "drink":
+            df = df[df["category_clean"].astype(str).apply(lambda x: normalize_category(x) in ("cafe", "restaurant") or "كافيه" in x or "مطعم" in x)]
+        else:
+            # Unknown item: scan likely menu-owning categories only, not pharmacy/supermarket/housing.
+            df = df[df["category_clean"].astype(str).apply(lambda x: normalize_category(x) in ("cafe", "restaurant") or "كافيه" in x or "مطعم" in x)]
+
+    matches = []
+    for _, row in df.head(max_places_to_scan).iterrows():
+        place = row.to_dict()
+        pid = _place_id_for_menu(place)
+        if not pid:
+            continue
+        menu_items = _get_menu_cached(pid)
+        matched_items = []
+        for item in menu_items:
+            blob = clean_text(
+                f"{item.get('item_name','')} {item.get('name','')} "
+                f"{item.get('subcategory_name','')} {item.get('sub_category','')} "
+                f"{item.get('description','')}"
+            )
+            if _text_matches_item_query(blob, item_query):
+                matched_items.append(item)
+        if matched_items:
+            place["matched_items"] = matched_items[:5]
+            place["matched_item"] = matched_items[0].get("item_name") or matched_items[0].get("name") or item_query
+            place["matched_item_count"] = len(matched_items)
+            matches.append(place)
+            if len(matches) >= CFG.MAX_CARDS:
+                break
+
+    log.info(f"🔎 scan_menus_for_item '{item_query}' → {len(matches)} places")
+    return matches
+
 # ════════════════════════════════════════════════════════════════
 # ─── RESPONSE FORMATTERS ────────────────────────────────────────
 # ════════════════════════════════════════════��═══════════════════
@@ -2470,6 +2614,11 @@ def handle_search_by_item(text: str, session, user_lat=None, user_lon=None) -> d
     places = search_places_by_item(item_query)
     places = filter_item_results_strict(places, item_query)
 
+    # If the dedicated endpoint/fuzzy search missed an item that actually exists
+    # in menus, scan real menus before saying no_result.
+    if not places:
+        places = scan_menus_for_item(item_query)
+
     if not places:
         result = {
             "reply":       f"مش لاقي أماكن بتقدم {item_query} حاليًا 😔",
@@ -2845,3 +2994,4 @@ def reset_session(session_id: str):
     SESSIONS.pop(session_id, None)
     log.info(f"🗑️  Session {session_id} reset")
     return {"status": "reset", "session_id": session_id}
+