feat: AI 업셀링 CRM - Clawdbot Gateway 기반 맞춤 추천 시스템

키오스크 적립 시 Clawdbot Gateway(Claude Max)를 통해 구매 이력 기반
맞춤 제품 추천을 생성하고, 마이페이지 방문 시 바텀시트 팝업으로 표시.

- ai_recommendations SQLite 테이블 추가 (스키마 + 마이그레이션)
- clawdbot_client.py: Gateway WebSocket 프로토콜 v3 Python 클라이언트
- app.py: 추천 생성 + GET/POST API 엔드포인트
- my_page.html: 바텀시트 UI (슬라이드업 애니메이션, 1.5초 후 자동 표시)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

backend/app.py

@@ -1109,7 +1109,7 @@ def my_page():
         tx_dict['created_at'] = utc_to_kst_str(tx['created_at'])
         transactions.append(tx_dict)
 
-    return render_template('my_page.html', user=user, transactions=transactions)
+    return render_template('my_page.html', user=user, transactions=transactions, user_id=user['id'])
 
 
 # ============================================================================
@@ -1885,6 +1885,137 @@ def admin():
                            recent_tokens=recent_tokens)
 
 
+# ============================================================================
+# AI 업셀링 추천
+# ============================================================================
+
+def _generate_upsell_recommendation(user_id, transaction_id, sale_items, user_name):
+    """키오스크 적립 후 AI 업셀링 추천 생성 (fire-and-forget)"""
+    from services.clawdbot_client import generate_upsell
+
+    if not sale_items:
+        return
+
+    # 현재 구매 품목
+    current_items = ', '.join(item['name'] for item in sale_items if item.get('name'))
+    if not current_items:
+        return
+
+    # 최근 구매 이력 수집
+    recent_products = current_items  # 기본값
+    try:
+        conn = db_manager.get_sqlite_connection()
+        cursor = conn.cursor()
+        cursor.execute("""
+            SELECT ct.transaction_id
+            FROM claim_tokens ct
+            WHERE ct.claimed_by_user_id = ? AND ct.transaction_id != ?
+            ORDER BY ct.claimed_at DESC LIMIT 5
+        """, (user_id, transaction_id))
+        recent_tokens = cursor.fetchall()
+
+        if recent_tokens:
+            all_products = []
+            mssql_session = db_manager.get_session('PM_PRES')
+            for token in recent_tokens:
+                rows = mssql_session.execute(text("""
+                    SELECT ISNULL(G.GoodsName, '') AS goods_name
+                    FROM SALE_SUB S
+                    LEFT JOIN PM_DRUG.dbo.CD_GOODS G ON S.DrugCode = G.DrugCode
+                    WHERE S.SL_NO_order = :tid
+                """), {'tid': token['transaction_id']}).fetchall()
+                for r in rows:
+                    if r.goods_name:
+                        all_products.append(r.goods_name)
+            if all_products:
+                recent_products = ', '.join(set(all_products))
+    except Exception as e:
+        logging.warning(f"[AI추천] 구매 이력 수집 실패 (현재 품목만 사용): {e}")
+
+    # Claude로 추천 생성
+    logging.info(f"[AI추천] 생성 시작: user={user_name}, items={current_items}")
+    rec = generate_upsell(user_name, current_items, recent_products)
+
+    if not rec:
+        logging.warning("[AI추천] 생성 실패 (AI 응답 없음)")
+        return
+
+    # SQLite에 저장
+    try:
+        conn = db_manager.get_sqlite_connection()
+        cursor = conn.cursor()
+        expires_at = (datetime.now() + timedelta(days=7)).strftime('%Y-%m-%d %H:%M:%S')
+        cursor.execute("""
+            INSERT INTO ai_recommendations
+            (user_id, transaction_id, recommended_product, recommendation_message,
+             recommendation_reason, trigger_products, ai_raw_response, expires_at)
+            VALUES (?, ?, ?, ?, ?, ?, ?, ?)
+        """, (
+            user_id, transaction_id,
+            rec['product'], rec['message'], rec['reason'],
+            json.dumps([item['name'] for item in sale_items], ensure_ascii=False),
+            json.dumps(rec, ensure_ascii=False),
+            expires_at
+        ))
+        conn.commit()
+        logging.info(f"[AI추천] 저장 완료: user_id={user_id}, product={rec['product']}")
+    except Exception as e:
+        logging.warning(f"[AI추천] DB 저장 실패: {e}")
+
+
+@app.route('/api/recommendation/<int:user_id>')
+def api_get_recommendation(user_id):
+    """마이페이지용 AI 추천 조회"""
+    conn = db_manager.get_sqlite_connection()
+    cursor = conn.cursor()
+
+    now = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
+    cursor.execute("""
+        SELECT id, recommended_product, recommendation_message, created_at
+        FROM ai_recommendations
+        WHERE user_id = ? AND status = 'active'
+          AND (expires_at IS NULL OR expires_at > ?)
+        ORDER BY created_at DESC LIMIT 1
+    """, (user_id, now))
+
+    rec = cursor.fetchone()
+    if not rec:
+        return jsonify({'success': True, 'has_recommendation': False})
+
+    # 표시 횟수 업데이트
+    cursor.execute("""
+        UPDATE ai_recommendations
+        SET displayed_count = displayed_count + 1,
+            displayed_at = COALESCE(displayed_at, ?)
+        WHERE id = ?
+    """, (now, rec['id']))
+    conn.commit()
+
+    return jsonify({
+        'success': True,
+        'has_recommendation': True,
+        'recommendation': {
+            'id': rec['id'],
+            'product': rec['recommended_product'],
+            'message': rec['recommendation_message']
+        }
+    })
+
+
+@app.route('/api/recommendation/<int:rec_id>/dismiss', methods=['POST'])
+def api_dismiss_recommendation(rec_id):
+    """추천 닫기"""
+    conn = db_manager.get_sqlite_connection()
+    cursor = conn.cursor()
+    now = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
+    cursor.execute("""
+        UPDATE ai_recommendations SET status = 'dismissed', dismissed_at = ?
+        WHERE id = ?
+    """, (now, rec_id))
+    conn.commit()
+    return jsonify({'success': True})
+
+
 # ============================================================================
 # 알림톡 로그
 # ============================================================================
@@ -2195,6 +2326,12 @@ def api_kiosk_claim():
         except Exception as alimtalk_err:
             logging.warning(f"[알림톡] 발송 예외 (적립은 완료): {alimtalk_err}")
 
+        # AI 업셀링 추천 생성 (fire-and-forget)
+        try:
+            _generate_upsell_recommendation(user_id, transaction_id, sale_items, user_name)
+        except Exception as rec_err:
+            logging.warning(f"[AI추천] 생성 예외 (적립은 완료): {rec_err}")
+
         return jsonify({
             'success': True,
             'message': f'{claimed_points}P 적립 완료!',

backend/db/dbsetup.py

@@ -269,6 +269,33 @@ class DatabaseManager:
             self.sqlite_conn.commit()
             print("[DB Manager] SQLite 마이그레이션: alimtalk_logs 테이블 생성")
 
+        # ai_recommendations 테이블 생성
+        cursor.execute("SELECT name FROM sqlite_master WHERE type='table' AND name='ai_recommendations'")
+        if not cursor.fetchone():
+            cursor.executescript("""
+                CREATE TABLE IF NOT EXISTS ai_recommendations (
+                    id INTEGER PRIMARY KEY AUTOINCREMENT,
+                    user_id INTEGER NOT NULL,
+                    transaction_id VARCHAR(20),
+                    recommended_product TEXT NOT NULL,
+                    recommendation_message TEXT NOT NULL,
+                    recommendation_reason TEXT,
+                    trigger_products TEXT,
+                    ai_raw_response TEXT,
+                    status VARCHAR(20) DEFAULT 'active',
+                    displayed_count INTEGER DEFAULT 0,
+                    created_at DATETIME DEFAULT CURRENT_TIMESTAMP,
+                    expires_at DATETIME,
+                    displayed_at DATETIME,
+                    dismissed_at DATETIME,
+                    FOREIGN KEY (user_id) REFERENCES users(id)
+                );
+                CREATE INDEX IF NOT EXISTS idx_rec_user_status ON ai_recommendations(user_id, status);
+                CREATE INDEX IF NOT EXISTS idx_rec_expires ON ai_recommendations(expires_at);
+            """)
+            self.sqlite_conn.commit()
+            print("[DB Manager] SQLite 마이그레이션: ai_recommendations 테이블 생성")
+
     def test_connection(self, database='PM_BASE'):
         """연결 테스트"""
         try:

backend/db/mileage_schema.sql

@@ -98,3 +98,25 @@ CREATE TABLE IF NOT EXISTS alimtalk_logs (
 
 CREATE INDEX IF NOT EXISTS idx_alimtalk_created ON alimtalk_logs(created_at);
 CREATE INDEX IF NOT EXISTS idx_alimtalk_recipient ON alimtalk_logs(recipient_no);
+
+-- 7. AI 추천 테이블
+CREATE TABLE IF NOT EXISTS ai_recommendations (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    user_id INTEGER NOT NULL,
+    transaction_id VARCHAR(20),
+    recommended_product TEXT NOT NULL,
+    recommendation_message TEXT NOT NULL,
+    recommendation_reason TEXT,
+    trigger_products TEXT,
+    ai_raw_response TEXT,
+    status VARCHAR(20) DEFAULT 'active',
+    displayed_count INTEGER DEFAULT 0,
+    created_at DATETIME DEFAULT CURRENT_TIMESTAMP,
+    expires_at DATETIME,
+    displayed_at DATETIME,
+    dismissed_at DATETIME,
+    FOREIGN KEY (user_id) REFERENCES users(id)
+);
+
+CREATE INDEX IF NOT EXISTS idx_rec_user_status ON ai_recommendations(user_id, status);
+CREATE INDEX IF NOT EXISTS idx_rec_expires ON ai_recommendations(expires_at);

backend/services/clawdbot_client.py

@@ -0,0 +1,268 @@
+"""
+Clawdbot Gateway Python 클라이언트
+카카오톡 봇과 동일한 Gateway WebSocket API를 통해 Claude와 통신
+추가 API 비용 없음 (Claude Max 구독 재활용)
+"""
+
+import json
+import uuid
+import asyncio
+import logging
+from pathlib import Path
+
+import websockets
+
+logger = logging.getLogger(__name__)
+
+# Gateway 설정 (clawdbot.json에서 읽기)
+CLAWDBOT_CONFIG_PATH = Path.home() / '.clawdbot' / 'clawdbot.json'
+
+
+def _load_gateway_config():
+    """clawdbot.json에서 Gateway 설정 로드"""
+    try:
+        with open(CLAWDBOT_CONFIG_PATH, 'r', encoding='utf-8') as f:
+            config = json.load(f)
+        gw = config.get('gateway', {})
+        return {
+            'port': gw.get('port', 18789),
+            'token': gw.get('auth', {}).get('token', ''),
+        }
+    except Exception as e:
+        logger.warning(f"[Clawdbot] 설정 파일 로드 실패: {e}")
+        return {'port': 18789, 'token': ''}
+
+
+async def _ask_gateway(message, session_id='pharmacy-upsell',
+                       system_prompt=None, timeout=60):
+    """
+    Clawdbot Gateway WebSocket API 호출
+
+    프로토콜:
+    1. WS 연결
+    2. 서버 → connect.challenge (nonce)
+    3. 클라이언트 → connect 요청 (token)
+    4. 서버 → connect 응답 (ok)
+    5. 클라이언트 → agent 요청
+    6. 서버 → accepted (ack) → 최종 응답
+
+    Returns:
+        str: AI 응답 텍스트 (실패 시 None)
+    """
+    config = _load_gateway_config()
+    url = f"ws://127.0.0.1:{config['port']}"
+    token = config['token']
+
+    try:
+        async with websockets.connect(url, max_size=25 * 1024 * 1024,
+                                       close_timeout=5) as ws:
+            # 1. connect.challenge 대기
+            nonce = None
+            challenge_msg = await asyncio.wait_for(ws.recv(), timeout=10)
+            challenge = json.loads(challenge_msg)
+            if challenge.get('event') == 'connect.challenge':
+                nonce = challenge.get('payload', {}).get('nonce')
+
+            # 2. connect 요청
+            connect_id = str(uuid.uuid4())
+            connect_frame = {
+                'type': 'req',
+                'id': connect_id,
+                'method': 'connect',
+                'params': {
+                    'minProtocol': 3,
+                    'maxProtocol': 3,
+                    'client': {
+                        'id': 'gateway-client',
+                        'displayName': 'Pharmacy Upsell',
+                        'version': '1.0.0',
+                        'platform': 'win32',
+                        'mode': 'backend',
+                        'instanceId': str(uuid.uuid4()),
+                    },
+                    'caps': [],
+                    'auth': {
+                        'token': token,
+                    },
+                    'role': 'operator',
+                    'scopes': ['operator.admin'],
+                }
+            }
+            await ws.send(json.dumps(connect_frame))
+
+            # 3. connect 응답 대기
+            while True:
+                msg = await asyncio.wait_for(ws.recv(), timeout=10)
+                data = json.loads(msg)
+                if data.get('id') == connect_id:
+                    if not data.get('ok'):
+                        error = data.get('error', {}).get('message', 'unknown')
+                        logger.warning(f"[Clawdbot] connect 실패: {error}")
+                        return None
+                    break  # 연결 성공
+
+            # 4. agent 요청
+            agent_id = str(uuid.uuid4())
+            agent_params = {
+                'message': message,
+                'sessionId': session_id,
+                'sessionKey': session_id,
+                'timeout': timeout,
+                'idempotencyKey': str(uuid.uuid4()),
+            }
+            if system_prompt:
+                agent_params['extraSystemPrompt'] = system_prompt
+
+            agent_frame = {
+                'type': 'req',
+                'id': agent_id,
+                'method': 'agent',
+                'params': agent_params,
+            }
+            await ws.send(json.dumps(agent_frame))
+
+            # 5. agent 응답 대기 (accepted → final)
+            while True:
+                msg = await asyncio.wait_for(ws.recv(), timeout=timeout + 30)
+                data = json.loads(msg)
+
+                # 이벤트 무시 (tick 등)
+                if data.get('event'):
+                    continue
+
+                # 우리 요청에 대한 응답인지 확인
+                if data.get('id') != agent_id:
+                    continue
+
+                payload = data.get('payload', {})
+                status = payload.get('status')
+
+                # accepted는 대기
+                if status == 'accepted':
+                    continue
+
+                # 최종 응답
+                if data.get('ok'):
+                    payloads = payload.get('result', {}).get('payloads', [])
+                    text = '\n'.join(p.get('text', '') for p in payloads if p.get('text'))
+                    return text or None
+                else:
+                    error = data.get('error', {}).get('message', 'unknown')
+                    logger.warning(f"[Clawdbot] agent 실패: {error}")
+                    return None
+
+    except asyncio.TimeoutError:
+        logger.warning("[Clawdbot] Gateway 타임아웃")
+        return None
+    except (ConnectionRefusedError, OSError) as e:
+        logger.warning(f"[Clawdbot] Gateway 연결 실패 (꺼져있음?): {e}")
+        return None
+    except Exception as e:
+        logger.warning(f"[Clawdbot] Gateway 오류: {e}")
+        return None
+
+
+def ask_clawdbot(message, session_id='pharmacy-upsell',
+                 system_prompt=None, timeout=60):
+    """
+    동기 래퍼: Flask에서 직접 호출 가능
+
+    Args:
+        message: 사용자 메시지
+        session_id: 세션 ID (대화 구분용)
+        system_prompt: 추가 시스템 프롬프트
+        timeout: 타임아웃 (초)
+
+    Returns:
+        str: AI 응답 텍스트 (실패 시 None)
+    """
+    try:
+        loop = asyncio.new_event_loop()
+        result = loop.run_until_complete(
+            _ask_gateway(message, session_id, system_prompt, timeout)
+        )
+        loop.close()
+        return result
+    except Exception as e:
+        logger.warning(f"[Clawdbot] 호출 실패: {e}")
+        return None
+
+
+# 업셀링 전용 ──────────────────────────────────────
+
+UPSELL_SYSTEM_PROMPT = """당신은 동네 약국(청춘약국)의 친절한 약사입니다.
+고객의 구매 이력을 보고, 자연스럽고 따뜻한 톤으로 약 하나를 추천합니다.
+강압적이거나 광고 같은 느낌이 아닌, 진심으로 건강을 걱정하는 약사의 말투로 작성해주세요.
+반드시 아래 JSON 형식으로만 응답하세요. 다른 텍스트 없이 JSON만 출력하세요."""
+
+
+def generate_upsell(user_name, current_items, recent_products):
+    """
+    업셀링 추천 생성
+
+    Args:
+        user_name: 고객명
+        current_items: 오늘 구매 품목 문자열 (예: "타이레놀, 챔프 시럽")
+        recent_products: 최근 구매 이력 문자열
+
+    Returns:
+        dict: {'product': '...', 'reason': '...', 'message': '...'} 또는 None
+    """
+    prompt = f"""고객 이름: {user_name}
+오늘 구매한 약: {current_items}
+최근 구매 이력: {recent_products}
+
+위 정보를 바탕으로 이 고객에게 추천할 약품 하나를 제안해주세요.
+
+규칙:
+1. 오늘 구매한 약과 함께 먹으면 좋거나, 구매 패턴상 필요해보이는 약 1가지만 추천
+2. 실제 약국에서 판매하는 일반의약품/건강기능식품만 추천 (처방약 제외)
+3. 메시지는 2문장 이내, 따뜻하고 자연스러운 톤
+4. 구체적인 제품명 사용 (예: "비타민C 1000", "오메가3" 등)
+
+응답은 반드시 아래 JSON 형식으로만:
+{{"product": "추천 제품명", "reason": "추천 이유 (내부용, 1문장)", "message": "{user_name}님, [오늘 구매 품목]과 함께 [추천약]도 추천드려요. [간단한 이유]."}}"""
+
+    response_text = ask_clawdbot(
+        prompt,
+        session_id=f'upsell-{user_name}',
+        system_prompt=UPSELL_SYSTEM_PROMPT,
+        timeout=30
+    )
+
+    if not response_text:
+        return None
+
+    return _parse_upsell_response(response_text)
+
+
+def _parse_upsell_response(text):
+    """AI 응답에서 JSON 추출"""
+    import re
+    try:
+        # ```json ... ``` 블록 추출 시도
+        json_match = re.search(r'```json\s*(\{.*?\})\s*```', text, re.DOTALL)
+        if json_match:
+            json_str = json_match.group(1)
+        else:
+            # 직접 JSON 파싱 시도
+            start = text.find('{')
+            end = text.rfind('}')
+            if start >= 0 and end > start:
+                json_str = text[start:end + 1]
+            else:
+                return None
+
+        data = json.loads(json_str)
+
+        if 'product' not in data or 'message' not in data:
+            return None
+
+        return {
+            'product': data['product'],
+            'reason': data.get('reason', ''),
+            'message': data['message'],
+        }
+    except (json.JSONDecodeError, Exception) as e:
+        logger.warning(f"[Clawdbot] 업셀 응답 파싱 실패: {e}")
+        return None

backend/templates/my_page.html

@@ -392,6 +392,57 @@
         }
     }
     </script>
+
+    <!-- AI 추천 바텀시트 -->
+    <div id="rec-sheet" style="display:none;">
+        <div id="rec-backdrop" style="position:fixed;top:0;left:0;right:0;bottom:0;background:rgba(0,0,0,0.3);z-index:999;animation:recFadeIn .3s ease;"></div>
+        <div id="rec-content" style="position:fixed;bottom:0;left:50%;transform:translateX(-50%);width:100%;max-width:420px;background:#fff;border-radius:24px 24px 0 0;padding:12px 24px 32px;box-shadow:0 -8px 32px rgba(0,0,0,0.12);z-index:1000;animation:recSlideUp .4s cubic-bezier(.16,1,.3,1);">
+            <div style="width:40px;height:4px;background:#dee2e6;border-radius:2px;margin:0 auto 20px;"></div>
+            <div style="text-align:center;padding:8px 0 20px;">
+                <div style="font-size:48px;margin-bottom:16px;">💊</div>
+                <div id="rec-message" style="color:#343a40;font-size:16px;font-weight:500;line-height:1.6;letter-spacing:-0.3px;margin-bottom:16px;"></div>
+                <div id="rec-product" style="display:inline-block;background:linear-gradient(135deg,#6366f1,#8b5cf6);color:#fff;font-size:14px;font-weight:600;padding:8px 20px;border-radius:20px;letter-spacing:-0.2px;"></div>
+            </div>
+            <div style="display:flex;gap:12px;padding-bottom:env(safe-area-inset-bottom,0);">
+                <button onclick="dismissRec()" style="flex:1;padding:14px;border:1px solid #dee2e6;border-radius:14px;background:#fff;color:#868e96;font-size:15px;font-weight:600;cursor:pointer;font-family:inherit;">다음에요</button>
+                <button onclick="dismissRec()" style="flex:2;padding:14px;border:none;border-radius:14px;background:linear-gradient(135deg,#6366f1,#8b5cf6);color:#fff;font-size:15px;font-weight:600;cursor:pointer;font-family:inherit;">관심있어요!</button>
+            </div>
+        </div>
+    </div>
+    <style>
+        @keyframes recFadeIn { from{opacity:0} to{opacity:1} }
+        @keyframes recSlideUp { from{transform:translate(-50%,100%)} to{transform:translate(-50%,0)} }
+        @keyframes recSlideDown { from{transform:translate(-50%,0)} to{transform:translate(-50%,100%)} }
+    </style>
+    <script>
+    let _recId = null;
+    window.addEventListener('load', function() {
+        {% if user_id %}
+        setTimeout(async function() {
+            try {
+                const res = await fetch('/api/recommendation/{{ user_id }}');
+                const data = await res.json();
+                if (data.success && data.has_recommendation) {
+                    _recId = data.recommendation.id;
+                    document.getElementById('rec-message').textContent = data.recommendation.message;
+                    document.getElementById('rec-product').textContent = data.recommendation.product;
+                    document.getElementById('rec-sheet').style.display = 'block';
+                    document.getElementById('rec-backdrop').onclick = dismissRec;
+                }
+            } catch(e) {}
+        }, 1500);
+        {% endif %}
+    });
+    function dismissRec() {
+        const c = document.getElementById('rec-content');
+        const b = document.getElementById('rec-backdrop');
+        c.style.animation = 'recSlideDown .3s ease forwards';
+        b.style.opacity = '0';
+        b.style.transition = 'opacity .3s';
+        setTimeout(function(){ document.getElementById('rec-sheet').style.display='none'; }, 300);
+        if (_recId) fetch('/api/recommendation/' + _recId + '/dismiss', {method:'POST'}).catch(function(){});
+    }
+    </script>
     <script>if('serviceWorker' in navigator){navigator.serviceWorker.register('/sw.js').catch(()=>{});}</script>
 </body>
 </html>