feat(ai): add rolling conversation summary memory

2026-04-13 11:59:43 +05:00
parent 989bf9a0c1
commit 6b8f5d069b
14 changed files with 900 additions and 26 deletions
@@ -0,0 +1,99 @@
+import json
+
+from services.shared.ai_context_summary import (
+    render_context_summary_text,
+    update_context_summary_from_assistant_turn,
+    update_context_summary_from_user_turn,
+)
+
+
+def test_context_summary_tracks_city_and_branch_slot_for_schedule():
+    summary = update_context_summary_from_user_turn(
+        None,
+        channel="voice",
+        language="ru",
+        customer_name="Ернур",
+        text="Мне нужен график работы в городе Алмата.",
+        now="2026-04-12T00:00:00+00:00",
+    )
+
+    assert summary["channel"] == "voice"
+    assert summary["customer_name"] == "Ернур"
+    assert summary["active_intent"] == "schedule"
+    assert summary["confirmed_facts"]["city"] == "Алмата"
+    assert summary["open_slots"] == ["branch_or_address"]
+
+
+def test_context_summary_does_not_overwrite_meaningful_request_with_low_signal():
+    summary = update_context_summary_from_user_turn(
+        None,
+        channel="voice",
+        language="ru",
+        customer_name="Ернур",
+        text="Мне нужен график работы в городе Алмата.",
+        now="2026-04-12T00:00:00+00:00",
+    )
+
+    updated = update_context_summary_from_user_turn(
+        json.dumps(summary, ensure_ascii=False),
+        channel="voice",
+        language="ru",
+        customer_name="Ернур",
+        text="Алло",
+        now="2026-04-12T00:00:02+00:00",
+    )
+
+    assert updated["active_request_text"] == "Мне нужен график работы в городе Алмата."
+    assert updated["confirmed_facts"]["city"] == "Алмата"
+
+
+def test_context_summary_assistant_turn_keeps_slots_and_renders_text():
+    summary = update_context_summary_from_user_turn(
+        None,
+        channel="telegram",
+        language="ru",
+        customer_name="Клиент",
+        text="Хочу узнать график работы в городе Алмата",
+        now="2026-04-12T00:00:00+00:00",
+    )
+
+    updated = update_context_summary_from_assistant_turn(
+        json.dumps(summary, ensure_ascii=False),
+        language="ru",
+        customer_name="Клиент",
+        reply_text="Подскажите, какой филиал или адрес в этом городе вас интересует?",
+        decision_intent="clarification",
+        needs_handoff=False,
+        handoff_reason=None,
+        now="2026-04-12T00:00:03+00:00",
+    )
+
+    rendered = render_context_summary_text(updated)
+    assert updated["open_slots"] == ["branch_or_address"]
+    assert "city=Алмата" in rendered
+    assert "график работы" in rendered
+
+
+def test_context_summary_generic_clarification_does_not_downgrade_specific_slot():
+    summary = update_context_summary_from_user_turn(
+        None,
+        channel="telegram",
+        language="ru",
+        customer_name="Клиент",
+        text="Хочу узнать график работы в городе Алмата",
+        now="2026-04-12T00:00:00+00:00",
+    )
+
+    updated = update_context_summary_from_assistant_turn(
+        json.dumps(summary, ensure_ascii=False),
+        language="ru",
+        customer_name="Клиент",
+        reply_text="Уточните, пожалуйста, что именно нужно проверить или подсказать.",
+        decision_intent="clarification",
+        needs_handoff=False,
+        handoff_reason=None,
+        now="2026-04-12T00:00:03+00:00",
+    )
+
+    assert updated["open_slots"] == ["branch_or_address"]
+    assert updated["requested_clarifications"] == ["branch_or_address"]
@@ -1590,6 +1590,71 @@ def test_turn_voice_session_early_plan_does_not_persist_partial_turns(monkeypatc
        session.close()


+def test_voice_llm_prompt_includes_context_summary_and_uses_12_segments():
+    transcript_window = [
+        SimpleNamespace(
+            speaker="caller" if index % 2 == 0 else "assistant",
+            text=f"segment {index}",
+            sequence_no=index,
+            source_type="voice_policy",
+            barge_in_interrupted=False,
+            created_at=utc_now_iso(),
+        )
+        for index in range(1, 16)
+    ]
+
+    messages = voice_module._voice_llm_prompt_messages(
+        language="ru",
+        customer=None,
+        interaction=SimpleNamespace(
+            interaction_id="int_voice_prompt",
+            status="open",
+            queue_id="que_voice",
+            subject="schedule",
+            customer_id="cus_voice_prompt",
+        ),
+        transcript_text="Мне нужен график работы",
+        transcript_window=transcript_window,
+        conversation_summary_text="Customer name: Ернур | Active intent: schedule | Confirmed facts: city=Алмата",
+        kb_results=[],
+        name_value="Ернур",
+        name_status="name_obtained",
+    )
+
+    payload = json.loads(messages[1]["content"])
+    assert payload["conversation_summary"].startswith("Customer name: Ернур")
+    assert len(payload["history"]) == 12
+    assert payload["history"][0]["sequence_no"] == 4
+
+
+def test_voice_postprocess_reply_uses_summary_context_when_raw_window_lost_topic():
+    reply_text = voice_module._voice_postprocess_reply_text(
+        language="ru",
+        transcript_text="Сколько раз повторять тебе?",
+        transcript_window=[
+            SimpleNamespace(speaker="caller", text="Сколько раз повторять тебе?", sequence_no=1, source_type="voice_asr", barge_in_interrupted=False, created_at=utc_now_iso()),
+        ],
+        context_summary=json.dumps(
+            {
+                "customer_name": "Ернур",
+                "active_intent": "schedule",
+                "active_request_text": "Мне нужен график работы в городе Алмата",
+                "confirmed_facts": {"city": "Алмата", "branch_hint": None, "service_hint": "график работы", "request_number": None},
+                "open_slots": ["branch_or_address"],
+            },
+            ensure_ascii=False,
+        ),
+        reply_text="Пожалуйста, уточните, какая услуга вас интересует, чтобы я мог подсказать тариф.",
+        kb_results=[],
+        needs_handoff=False,
+    )
+
+    normalized = reply_text.lower()
+    assert "тариф" not in normalized
+    assert "услуг" not in normalized
+    assert "филиал" in normalized or "адрес" in normalized
+
+
 def test_ai_enqueue_creates_outbound_ai_reply_and_delivery_flow(monkeypatch):
    monkeypatch.setenv("AI_TELEGRAM_ENABLED", "1")
    monkeypatch.setenv("AI_PROVIDER", "stub")
@@ -1639,6 +1704,61 @@ def test_ai_enqueue_creates_outbound_ai_reply_and_delivery_flow(monkeypatch):
    assert delivered.json()[-1]["telegram_message_id_external"] == "5551"


+def test_ai_enqueue_persists_internal_context_summary(monkeypatch):
+    monkeypatch.setenv("AI_TELEGRAM_ENABLED", "1")
+    monkeypatch.setenv("AI_PROVIDER", "stub")
+    monkeypatch.setattr(telegram_module, "_ai_enqueue_request", lambda thread_id, trigger_message_id: None)
+    monkeypatch.setattr(telegram_module, "_start_telegram_reply_delivery", lambda message_id: None)
+
+    telegram_client = TestClient(telegram_app)
+    interaction_client = TestClient(interaction_app)
+    ai_client = TestClient(ai_app)
+    patch_ai_internal_calls(monkeypatch, telegram_client, interaction_client)
+
+    created = create_inbound_thread(
+        telegram_client,
+        f"chat_ai_context_summary_{new_id('chat')}",
+        "Хочу узнать график работы в городе Алмата",
+    )
+    enqueue = ai_client.post(
+        f"/ai/telegram/threads/{created['thread_id']}/enqueue",
+        headers=admin_headers(),
+        json={"trigger_message_id": created["message_id"]},
+    )
+
+    assert enqueue.status_code == 200
+    assert enqueue.json()["status"] in {"done", "handoff_required", "running"}
+
+    deadline = time.time() + 2.0
+    last_summary = None
+    while time.time() < deadline:
+        session = get_session()
+        try:
+            thread = session.execute(
+                select(TelegramThreadRow).where(TelegramThreadRow.thread_id == created["thread_id"])
+            ).scalar_one()
+            if not thread.ai_session_id:
+                time.sleep(0.05)
+                continue
+            ai_session = session.execute(
+                select(AISessionRow).where(AISessionRow.session_id == thread.ai_session_id)
+            ).scalar_one()
+            last_summary = json.loads(ai_session.context_summary_json or "{}")
+            if (
+                last_summary.get("active_intent") == "schedule"
+                and last_summary.get("confirmed_facts", {}).get("city") == "Алмата"
+            ):
+                break
+        finally:
+            session.close()
+        time.sleep(0.05)
+
+    assert last_summary is not None
+    assert last_summary["active_intent"] == "schedule"
+    assert last_summary["confirmed_facts"]["city"] == "Алмата"
+    assert last_summary["open_slots"] == ["branch_or_address"]
+
+
 def test_ai_enqueue_relaxed_kb_search_answers_phrase_query(monkeypatch):
    monkeypatch.setenv("AI_TELEGRAM_ENABLED", "1")
    monkeypatch.setenv("AI_PROVIDER", "stub")
@@ -64,3 +64,42 @@ def test_init_sql_schema_backfills_whatsapp_message_runtime_columns():
    assert "author_type" in columns
    assert "ix_whatsapp_messages_chat_external_unique" in indexes
    assert "idx_whatsapp_messages_next_delivery_attempt_at" in indexes
+
+
+def test_init_sql_schema_backfills_ai_session_context_summary_columns():
+    with engine.begin() as conn:
+        conn.execute(text("DROP TABLE IF EXISTS ai_sessions"))
+        conn.execute(
+            text(
+                """
+                CREATE TABLE ai_sessions (
+                  id INTEGER PRIMARY KEY AUTOINCREMENT,
+                  session_id VARCHAR(64) NOT NULL UNIQUE,
+                  channel VARCHAR(32) NOT NULL,
+                  thread_id VARCHAR(64) NULL,
+                  interaction_id VARCHAR(64) NULL,
+                  customer_id VARCHAR(64) NULL,
+                  agent_profile VARCHAR(64) NOT NULL,
+                  language VARCHAR(16) NULL,
+                  status VARCHAR(32) NOT NULL,
+                  summary_text TEXT NOT NULL,
+                  last_user_message_id VARCHAR(64) NULL,
+                  last_ai_message_id VARCHAR(64) NULL,
+                  handoff_reason TEXT NULL,
+                  created_at VARCHAR(64) NOT NULL,
+                  updated_at VARCHAR(64) NOT NULL,
+                  closed_at VARCHAR(64) NULL
+                )
+                """
+            )
+        )
+
+    init_sql_schema()
+
+    inspector = inspect(engine)
+    columns = {item["name"] for item in inspector.get_columns("ai_sessions")}
+    indexes = {item["name"] for item in inspector.get_indexes("ai_sessions")}
+
+    assert "context_summary_json" in columns
+    assert "context_summary_updated_at" in columns
+    assert "idx_ai_sessions_context_summary_updated_at" in indexes