diff --git a/backend/ai_services/bailian_service.py b/backend/ai_services/bailian_service.py
index 62dc8f3..2c55725 100644
--- a/backend/ai_services/bailian_service.py
+++ b/backend/ai_services/bailian_service.py
@@ -95,12 +95,28 @@ class BailianService:
                 {'role': 'user', 'content': f"{prompt}\n\n以下是需要评估的内容：\n{content}{chapter_context}"}
             ]
 
-            completion = self.client.chat.completions.create(
-                model=evaluation.model_selection,
-                messages=messages,
-                response_format={"type": "json_object"} 
-            )
+            # 增加重试机制 (最多重试3次)
+            completion = None
+            last_error = None
+            import time
             
+            for attempt in range(3):
+                try:
+                    completion = self.client.chat.completions.create(
+                        model=evaluation.model_selection,
+                        messages=messages,
+                        response_format={"type": "json_object"} 
+                    )
+                    break # 成功则跳出循环
+                except Exception as e:
+                    last_error = e
+                    logger.warning(f"AI Evaluation attempt {attempt+1}/3 failed for eval {evaluation.id}: {e}")
+                    if attempt < 2:
+                        time.sleep(2 * (attempt + 1)) # 简单的指数退避
+            
+            if not completion:
+                raise last_error or Exception("AI Service call failed after retries")
+
             response_content = completion.choices[0].message.content
             # Convert to dict for storage
             raw_response = completion.model_dump()
diff --git a/backend/ai_services/views.py b/backend/ai_services/views.py
index 51a4c8d..cdadb9e 100644
--- a/backend/ai_services/views.py
+++ b/backend/ai_services/views.py
@@ -225,9 +225,34 @@ class TranscriptionTaskViewSet(viewsets.ModelViewSet):
                 eval.save()
                 evaluations_to_process.append(eval)
                 
-        # 执行评估
-        for eval_obj in evaluations_to_process:
-            service.evaluate_task(eval_obj)
+        # 执行评估 (改为异步并发执行)
+        # 提取ID列表，避免传递模型对象导致可能的线程问题
+        eval_ids = [e.id for e in evaluations_to_process]
+        
+        if eval_ids:
+            import threading
+            from concurrent.futures import ThreadPoolExecutor
+            
+            def run_evaluations_background(ids):
+                # 在后台线程中重新引入依赖
+                from .models import AIEvaluation
+                from .bailian_service import BailianService
+                
+                # 为该线程创建独立的服务实例
+                local_service = BailianService()
+                
+                # 获取最新的对象
+                target_evals = AIEvaluation.objects.filter(id__in=ids)
+                
+                # 使用线程池并发执行
+                # max_workers=4 可以同时处理4个评估请求
+                with ThreadPoolExecutor(max_workers=4) as executor:
+                    executor.map(local_service.evaluate_task, target_evals)
+            
+            # 启动后台线程，不阻塞当前 HTTP 请求
+            thread = threading.Thread(target=run_evaluations_background, args=(eval_ids,))
+            thread.daemon = True # 设置为守护线程
+            thread.start()
             
         # 返回该任务的所有评估结果
         all_evals = AIEvaluation.objects.filter(task=task)