Update app.py

app.py

@@ -35,10 +35,6 @@ def get_embedding(text: str) -> List[float]:
     )
     return response.data[0].embedding
 
-def format_vector_for_pg(vector: List[float]) -> str:
-    """벡터를 PostgreSQL 포맷으로 변환합니다."""
-    return f"[{','.join(str(x) for x in vector)}]"
-
 def get_text_value(node, field_name):
     """JSON 노드에서 텍스트 값을 안전하게 추출합니다."""
     if node and field_name in node and node[field_name] is not None:
@@ -70,23 +66,22 @@ def search_similar_chat(query: str, max_results: int = 100) -> List[Dict]:
     try:
         # 쿼리 임베딩 생성
         query_embedding = np.array(get_embedding(query))
-        query_vector = format_vector_for_pg(query_embedding)
         
         # DB 연결
         conn = get_db_conn()
         register_vector(conn)
         
-        # 여러 필드를 가중치로 조합한 유사도 검색 SQL
-        sql = f"""
+        # 여러 필드를 가중치로 조합한 유사도 검색 SQL - 매개변수화된 쿼리 사용
+        sql = """
             WITH embeddings AS (
                 SELECT 
                     id, 
                     metadata, 
                     content,
-                    CASE WHEN full_embedding IS NOT NULL THEN 1 - (full_embedding <=> '{query_vector}'::vector) ELSE 0 END * {full_w} as full_sim,
-                    CASE WHEN topic_embedding IS NOT NULL THEN 1 - (topic_embedding <=> '{query_vector}'::vector) ELSE 0 END * {topic_w} as topic_sim,
-                    CASE WHEN customer_embedding IS NOT NULL THEN 1 - (customer_embedding <=> '{query_vector}'::vector) ELSE 0 END * {customer_w} as customer_sim,
-                    CASE WHEN agent_embedding IS NOT NULL THEN 1 - (agent_embedding <=> '{query_vector}'::vector) ELSE 0 END * {agent_w} as agent_sim
+                    CASE WHEN full_embedding IS NOT NULL THEN 1 - (full_embedding <=> %s::vector) ELSE 0 END * %s as full_sim,
+                    CASE WHEN topic_embedding IS NOT NULL THEN 1 - (topic_embedding <=> %s::vector) ELSE 0 END * %s as topic_sim,
+                    CASE WHEN customer_embedding IS NOT NULL THEN 1 - (customer_embedding <=> %s::vector) ELSE 0 END * %s as customer_sim,
+                    CASE WHEN agent_embedding IS NOT NULL THEN 1 - (agent_embedding <=> %s::vector) ELSE 0 END * %s as agent_sim
                 FROM vector_store_multi_embeddings
                 WHERE full_embedding IS NOT NULL
                 OR topic_embedding IS NOT NULL
@@ -104,7 +99,15 @@ def search_similar_chat(query: str, max_results: int = 100) -> List[Dict]:
         """
         
         with conn.cursor() as cur:
-            cur.execute(sql, (limit,))
+            # 매개변수화된 쿼리 실행
+            params = (
+                query_embedding, full_w, 
+                query_embedding, topic_w, 
+                query_embedding, customer_w, 
+                query_embedding, agent_w, 
+                limit
+            )
+            cur.execute(sql, params)
             rows = cur.fetchall()
             
             results = []
@@ -204,23 +207,22 @@ def search_similar_chat_by_date(
         
         # 쿼리 임베딩 생성
         query_embedding = np.array(get_embedding(query))
-        query_vector = format_vector_for_pg(query_embedding)
         
         # DB 연결
         conn = get_db_conn()
         register_vector(conn)
         
-        # 여러 필드를 가중치로 조합한 유사도 검색 SQL (날짜 필터 추가)
-        sql = f"""
+        # 여러 필드를 가중치로 조합한 유사도 검색 SQL - 매개변수화
+        sql = """
             WITH embeddings AS (
                 SELECT 
                     id, 
                     metadata, 
                     content,
-                    CASE WHEN full_embedding IS NOT NULL THEN 1 - (full_embedding <=> '{query_vector}'::vector) ELSE 0 END * {full_w} as full_sim,
-                    CASE WHEN topic_embedding IS NOT NULL THEN 1 - (topic_embedding <=> '{query_vector}'::vector) ELSE 0 END * {topic_w} as topic_sim,
-                    CASE WHEN customer_embedding IS NOT NULL THEN 1 - (customer_embedding <=> '{query_vector}'::vector) ELSE 0 END * {customer_w} as customer_sim,
-                    CASE WHEN agent_embedding IS NOT NULL THEN 1 - (agent_embedding <=> '{query_vector}'::vector) ELSE 0 END * {agent_w} as agent_sim
+                    CASE WHEN full_embedding IS NOT NULL THEN 1 - (full_embedding <=> %s::vector) ELSE 0 END * %s as full_sim,
+                    CASE WHEN topic_embedding IS NOT NULL THEN 1 - (topic_embedding <=> %s::vector) ELSE 0 END * %s as topic_sim,
+                    CASE WHEN customer_embedding IS NOT NULL THEN 1 - (customer_embedding <=> %s::vector) ELSE 0 END * %s as customer_sim,
+                    CASE WHEN agent_embedding IS NOT NULL THEN 1 - (agent_embedding <=> %s::vector) ELSE 0 END * %s as agent_sim
                 FROM vector_store_multi_embeddings
                 WHERE full_embedding IS NOT NULL
                 OR topic_embedding IS NOT NULL
@@ -228,15 +230,20 @@ def search_similar_chat_by_date(
                 OR agent_embedding IS NOT NULL
         """
         
-        params = []
+        params = [
+            query_embedding, full_w, 
+            query_embedding, topic_w, 
+            query_embedding, customer_w, 
+            query_embedding, agent_w
+        ]
         
         # 날짜 필터 추가
         if start_timestamp is not None:
-            sql += f" AND (metadata->>'startTime')::bigint >= %s"
+            sql += " AND (metadata->>'startTime')::bigint >= %s"
             params.append(start_timestamp)
         
         if end_timestamp is not None:
-            sql += f" AND (metadata->>'startTime')::bigint <= %s"
+            sql += " AND (metadata->>'startTime')::bigint <= %s"
             params.append(end_timestamp)
         
         sql += """