fix: filter RU/KZ stopwords from KB search so filler words can't false-match

search_kb_rows had no relevance floor: any exact-token hit, however
generic, scored above zero and could win as the top/only result. A
caller utterance as thin as a bare "да" (confirming the language) could
exact-match that same common word inside an unrelated FAQ article's
body and get returned as "the" answer, which then got read back
almost verbatim — this is what surfaced live as the AI unprompted
launching into a voucher-activation explanation right after the
customer confirmed Russian, having said nothing else.

tokenize_kb_text now drops a curated set of RU/KZ greetings,
confirmations, pronouns, and particles. A stopword-only query naturally
falls through to the existing "no query tokens -> no results" path
instead of returning a coincidental match; genuine single-content-word
queries (e.g. "ваучер") are unaffected. Applies to every channel that
calls _kb_search (voice, Telegram, WhatsApp), not just voice.

services/shared/kb_search.py

@@ -19,6 +19,34 @@ _MIN_SOFT_MATCH_LENGTH = 4
 _NON_WORD_RE = re.compile(r"[^\w]+", re.UNICODE)
 _SPACE_RE = re.compile(r"\s+")
 
+# Common RU/KZ greetings, confirmations, pronouns, and particles that carry no
+# topical signal on their own. Without this, a caller utterance as thin as
+# "да" or "хорошо" could still exact-token-match some unrelated KB article
+# that happens to contain that word in its body, and get returned as the
+# top/only search result — read back to the caller as if it were the answer
+# to their question. Filtering these keeps _score_row's exact-token match
+# meaningful: a match now requires an actual content word.
+_STOPWORDS = frozenset(
+    {
+        # RU: greetings / confirmations / fillers
+        "алло", "ага", "да", "неа", "нет", "ой", "ок", "окей", "угу", "ясно",
+        "ладно", "хорошо", "понял", "поняла", "привет", "здравствуйте",
+        "добрый", "день", "вечер", "утро", "слышу", "слышно", "спасибо",
+        "пожалуйста", "извините", "простите", "алло",
+        # RU: pronouns / conjunctions / particles with no topical content
+        "я", "ты", "вы", "мы", "он", "она", "они", "это", "то", "и", "а",
+        "но", "или", "что", "как", "где", "когда", "если", "чтобы", "для",
+        "из", "по", "на", "в", "с", "у", "о", "же", "ли", "бы", "не", "ну",
+        "вот", "просто", "есть", "быть", "можно", "нужно", "надо", "уже",
+        # KZ: greetings / confirmations / fillers
+        "иә", "ия", "жоқ", "жарайды", "түсінікті", "рахмет", "сәлем",
+        "сәлеметсіз", "бе", "кешіріңіз",
+        # KZ: pronouns / conjunctions / particles
+        "мен", "сен", "сіз", "біз", "ол", "олар", "және", "бірақ", "немесе",
+        "не", "қалай", "қайда", "қашан", "үшін", "туралы",
+    }
+)
+
 
 class KBSearchRow(Protocol):
     id: int
@@ -38,7 +66,11 @@ def normalize_kb_text(value: str | None) -> str:
 
 
 def tokenize_kb_text(value: str | None) -> list[str]:
-    return [token for token in normalize_kb_text(value).split(" ") if len(token) >= _MIN_TOKEN_LENGTH]
+    return [
+        token
+        for token in normalize_kb_text(value).split(" ")
+        if len(token) >= _MIN_TOKEN_LENGTH and token not in _STOPWORDS
+    ]
 
 
 def search_kb_rows(

tests/test_kb_search.py

@@ -114,6 +114,61 @@ def test_search_kb_rows_prefers_title_and_tags_over_body_only_mentions():
     assert results[0].title == rows[0].title
 
 
+def test_search_kb_rows_ignores_stopword_only_query():
+    # A caller confirming the language ("да") should never surface an
+    # unrelated FAQ article just because that article's body happens to
+    # contain the word "да" somewhere in ordinary prose.
+    rows = [
+        _row(
+            row_id=1,
+            title="Активация ваучера",
+            body="Да, подтвердите СМС с номера 1414 командой 21*1.",
+            tags=["ваучер"],
+        ),
+    ]
+
+    assert search_kb_rows(rows, "да", limit=5) == []
+    assert search_kb_rows(rows, "Хорошо", limit=5) == []
+
+
+def test_search_kb_rows_still_matches_real_content_word_amid_fillers():
+    rows = [
+        _row(
+            row_id=1,
+            title="Активация ваучера",
+            body="Подтвердите СМС с номера 1414 командой 21*1.",
+            tags=["ваучер"],
+        ),
+        _row(
+            row_id=2,
+            title="График работы",
+            body="Филиалы работают с 9 до 18.",
+            tags=["график"],
+        ),
+    ]
+
+    results = search_kb_rows(rows, "да, у меня вопрос про ваучер", limit=5)
+
+    assert results
+    assert results[0].title == "Активация ваучера"
+
+
+def test_search_kb_rows_single_real_word_query_still_matches():
+    rows = [
+        _row(
+            row_id=1,
+            title="Активация ваучера",
+            body="Подтвердите СМС с номера 1414 командой 21*1.",
+            tags=["ваучер"],
+        ),
+    ]
+
+    results = search_kb_rows(rows, "ваучер", limit=5)
+
+    assert results
+    assert results[0].title == "Активация ваучера"
+
+
 def test_search_kb_rows_breaks_ties_by_newer_id():
     older = _row(
         row_id=10,