Update app.py

app.py

@@ -470,14 +470,2320 @@ class ResponseCleaner:
     @staticmethod
     def clean_response(response: str) -> str:
         """불필요한 마크업 제거 강화"""
+        # 마크다운 테이블 제거
+        # | 로 시작하는 테이블 행 제거
+        response = re.sub(r'^\|.*\|.*
+
+
+# ============================================================================
+# 통합 최적화 멀티 에이전트 시스템 (캐싱 제거 버전)
+# ============================================================================
+
+class SpeedOptimizedMultiAgentSystem:
+    """속도 최적화된 멀티 에이전트 시스템 (캐싱 없음)"""
+    
+    def __init__(self):
+        self.llm = OptimizedFireworksClient()
+        self.search = AsyncBraveSearch()
+        self.reasoning = LightweightReasoningChain()
+        self.quality_checker = QualityChecker()
+        self.streaming = OptimizedStreaming()
+        self.language_detector = LanguageDetector()
+        self.response_cleaner = ResponseCleaner()
+        
+        # 병렬 처리 풀
+        self.executor = ThreadPoolExecutor(max_workers=4)
+    
+    def _init_compact_prompts(self, lang: str = 'ko') -> Dict:
+        """압축된 고효율 프롬프트 (언어별)"""
+        prompts = {
+            'ko': {
+                AgentRole.SUPERVISOR: """[감독자-구조설계]
+즉시분석: 핵심의도+필요정보+답변구조
+출력: 5개 핵심포인트(각 1문장)
+추론체계 명시""",
+                
+                AgentRole.CREATIVE: """[창의성생성자]
+입력구조 따라 창의적 확장
+실용예시+혁신접근+구체조언
+불필요설명 제거""",
+                
+                AgentRole.CRITIC: """[비평자-검증]
+신속검토: 정확성/논리성/실용성
+개선포인트 3개만
+각 2문장 이내""",
+                
+                AgentRole.FINALIZER: """[최종통합]
+모든의견 종합→최적답변
+명확구조+실용정보+창의균형
+바로 핵심 내용부터 시작. 불필요한 헤더나 마크업 없이. 마크다운 헤더(#, ##, ###) 사용 금지."""
+            },
+            'en': {
+                AgentRole.SUPERVISOR: """[Supervisor-Structure]
+Immediate analysis: core intent+required info+answer structure
+Output: 5 key points (1 sentence each)
+Clear reasoning framework""",
+                
+                AgentRole.CREATIVE: """[Creative Generator]
+Follow structure, expand creatively
+Practical examples+innovative approach+specific advice
+Remove unnecessary explanations""",
+                
+                AgentRole.CRITIC: """[Critic-Verification]
+Quick review: accuracy/logic/practicality
+Only 3 improvement points
+Max 2 sentences each""",
+                
+                AgentRole.FINALIZER: """[Final Integration]
+Synthesize all inputs→optimal answer
+Clear structure+practical info+creative balance
+Start with core content directly. No unnecessary headers or markup. No markdown headers (#, ##, ###)."""
+            },
+            'ja': {
+                AgentRole.SUPERVISOR: """[監督者-構造設計]
+即時分析：核心意図+必要情報+回答構造
+出力：5つの核心ポイント（各1文）
+推論体系明示""",
+                
+                AgentRole.CREATIVE: """[創造性生成者]
+入力構造に従って創造的拡張
+実用例+革新的アプローチ+具体的アドバイス
+不要な説明削除""",
+                
+                AgentRole.CRITIC: """[批評者-検証]
+迅速レビュー：正確性/論理性/実用性
+改善ポイント3つのみ
+各2文以内""",
+                
+                AgentRole.FINALIZER: """[最終統合]
+全意見統合→最適回答
+明確構造+実用情報+創造性バランス
+核心内容から直接開始。不要なヘッダーやマークアップなし。マークダウンヘッダー（#、##、###）使用禁止。"""
+            },
+            'zh': {
+                AgentRole.SUPERVISOR: """[主管-结构设计]
+立即分析：核心意图+所需信息+答案结构
+输出：5个核心要点（每个1句）
+推理体系明确""",
+                
+                AgentRole.CREATIVE: """[创意生成器]
+按结构创造性扩展
+实用示例+创新方法+具体建议
+删除不必要的解释""",
+                
+                AgentRole.CRITIC: """[评论家-验证]
+快速审查：准确性/逻辑性/实用性
+仅3个改进点
+每个最多2句""",
+                
+                AgentRole.FINALIZER: """[最终整合]
+综合所有意见→最佳答案
+清晰结构+实用信息+创意平衡
+直接从核心内容开始。无需不必要的标题或标记。禁止使用Markdown标题（#、##、###）。"""
+            }
+        }
+        
+        return prompts.get(lang, prompts['en'])
+    
+    async def parallel_process_agents(
+        self,
+        query: str,
+        search_results: List[Dict],
+        show_progress: bool = True,
+        lang: str = None
+    ) -> AsyncGenerator[Tuple[str, str], None]:
+        """병렬 처리 파이프라인 (캐싱 없음)"""
+        
+        start_time = time.time()
+        
+        # 언어 자동 감지
+        if lang is None:
+            lang = self.language_detector.detect_language(query)
+        
+        # 언어별 프롬프트 설정
+        self.compact_prompts = self._init_compact_prompts(lang)
+        
+        search_context = self._format_search_results(search_results)
+        accumulated_response = ""
+        agent_thoughts = ""
+        
+        # 추론 패턴 결정
+        reasoning_pattern = self.reasoning.get_reasoning_pattern(query, lang)
+        
+        try:
+            # === 1단계: 감독자 + 검색 병렬 실행 ===
+            if show_progress:
+                progress_msg = {
+                    'ko': "🚀 병렬 처리 시작\n👔 감독자 분석 + 🔍 추가 검색 동시 진행...\n\n",
+                    'en': "🚀 Starting parallel processing\n👔 Supervisor analysis + 🔍 Additional search in progress...\n\n",
+                    'ja': "🚀 並列処理開始\n👔 監督者分析 + 🔍 追加検索同時進行中...\n\n",
+                    'zh': "🚀 开始并行处理\n👔 主管分析 + 🔍 附加搜索同时进行...\n\n"
+                }
+                agent_thoughts = progress_msg.get(lang, progress_msg['en'])
+                yield accumulated_response, agent_thoughts
+            
+            # 감독자 프롬프트 (언어별)
+            supervisor_prompt_templates = {
+                'ko': f"""
+질문: {query}
+검색결과: {search_context}
+추론패턴: {reasoning_pattern}
+즉시 핵심구조 5개 제시""",
+                'en': f"""
+Question: {query}
+Search results: {search_context}
+Reasoning pattern: {reasoning_pattern}
+Immediately provide 5 key structures""",
+                'ja': f"""
+質問: {query}
+検索結果: {search_context}
+推論パターン: {reasoning_pattern}
+即座に5つの核心構造を提示""",
+                'zh': f"""
+问题: {query}
+搜索结果: {search_context}
+推理模式: {reasoning_pattern}
+立即提供5个核心结构"""
+            }
+            
+            supervisor_prompt = supervisor_prompt_templates.get(lang, supervisor_prompt_templates['en'])
+            
+            supervisor_response = ""
+            supervisor_task = self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.SUPERVISOR]},
+                    {"role": "user", "content": supervisor_prompt}
+                ],
+                temperature=0.3,
+                max_tokens=500
+            )
+            
+            # 감독자 스트리밍 (버퍼링)
+            async for chunk in self.streaming.buffer_and_yield(supervisor_task):
+                supervisor_response += chunk
+                if show_progress and len(supervisor_response) < 300:
+                    supervisor_label = {
+                        'ko': "👔 감독자 분석",
+                        'en': "👔 Supervisor Analysis",
+                        'ja': "👔 監督者分析",
+                        'zh': "👔 主管分析"
+                    }
+                    agent_thoughts = f"{supervisor_label.get(lang, supervisor_label['en'])}\n{supervisor_response[:300]}...\n\n"
+                    yield accumulated_response, agent_thoughts
+            
+            # === 2단계: 창의성 + 비평 준비 병렬 ===
+            if show_progress:
+                creative_msg = {
+                    'ko': "🎨 창의성 생성자 + 🔍 비평자 준비...\n\n",
+                    'en': "🎨 Creative Generator + 🔍 Critic preparing...\n\n",
+                    'ja': "🎨 創造性生成者 + 🔍 批評者準備中...\n\n",
+                    'zh': "🎨 创意生成器 + 🔍 评论家准备中...\n\n"
+                }
+                agent_thoughts += creative_msg.get(lang, creative_msg['en'])
+                yield accumulated_response, agent_thoughts
+            
+            # 창의성 생성 시작 (언어별)
+            creative_prompt_templates = {
+                'ko': f"""
+질문: {query}
+감독자구조: {supervisor_response}
+검색결과: {search_context}
+창의적+실용적 답변 즉시생성""",
+                'en': f"""
+Question: {query}
+Supervisor structure: {supervisor_response}
+Search results: {search_context}
+Generate creative+practical answer immediately""",
+                'ja': f"""
+質問: {query}
+監督者構造: {supervisor_response}
+検索結果: {search_context}
+創造的+実用的回答即座生成""",
+                'zh': f"""
+问题: {query}
+主管结构: {supervisor_response}
+搜索结果: {search_context}
+立即生成创意+实用答案"""
+            }
+            
+            creative_prompt = creative_prompt_templates.get(lang, creative_prompt_templates['en'])
+            
+            creative_response = ""
+            creative_partial = ""
+            critic_started = False
+            critic_response = ""
+            
+            creative_task = self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.CREATIVE]},
+                    {"role": "user", "content": creative_prompt}
+                ],
+                temperature=0.8,
+                max_tokens=1500
+            )
+            
+            # 창의성 스트리밍 + 비평자 조기 시작
+            async for chunk in self.streaming.buffer_and_yield(creative_task):
+                creative_response += chunk
+                creative_partial += chunk
+                
+                # 창의성 응답이 500자 넘으면 비평자 시작
+                if len(creative_partial) > 500 and not critic_started:
+                    critic_started = True
+                    
+                    # 비평자 비동기 시작 (언어별)
+                    critic_prompt_templates = {
+                        'ko': f"""
+원본질문: {query}
+창의성답변(일부): {creative_partial}
+신속검토→개선점3개""",
+                        'en': f"""
+Original question: {query}
+Creative answer (partial): {creative_partial}
+Quick review→3 improvements""",
+                        'ja': f"""
+元の質問: {query}
+創造的回答（一部）: {creative_partial}
+迅速レビュー→改善点3つ""",
+                        'zh': f"""
+原始问题: {query}
+创意答案（部分）: {creative_partial}
+快速审查→3个改进点"""
+                    }
+                    
+                    critic_prompt = critic_prompt_templates.get(lang, critic_prompt_templates['en'])
+                    
+                    critic_task = asyncio.create_task(
+                        self._run_critic_async(critic_prompt)
+                    )
+                
+                if show_progress:
+                    display_creative = creative_response[:400] + "..." if len(creative_response) > 400 else creative_response
+                    creative_label = {
+                        'ko': "🎨 창의성 생성자",
+                        'en': "🎨 Creative Generator",
+                        'ja': "🎨 創造性生成者",
+                        'zh': "🎨 创意生成器"
+                    }
+                    agent_thoughts = f"{creative_label.get(lang, creative_label['en'])}\n{display_creative}\n\n"
+                    yield accumulated_response, agent_thoughts
+            
+            # 비평자 결과 대기
+            if critic_started:
+                critic_response = await critic_task
+                
+                if show_progress:
+                    critic_label = {
+                        'ko': "🔍 비평자 검토",
+                        'en': "🔍 Critic Review",
+                        'ja': "🔍 批評者レビュー",
+                        'zh': "🔍 评论家审查"
+                    }
+                    agent_thoughts += f"{critic_label.get(lang, critic_label['en'])}\n{critic_response[:200]}...\n\n"
+                    yield accumulated_response, agent_thoughts
+            
+            # === 3단계: 품질 체크 및 조기 종료 ===
+            quality_score, need_more = self.quality_checker.evaluate_response(
+                creative_response, query, lang
+            )
+            
+            if not need_more and quality_score > 0.85:
+                # 품질이 충분히 높으면 바로 반환
+                accumulated_response = self.response_cleaner.clean_response(creative_response)
+                
+                if show_progress:
+                    quality_msg = {
+                        'ko': f"✅ 품질 충족 (점수: {quality_score:.2f})\n조기 완료!\n",
+                        'en': f"✅ Quality met (score: {quality_score:.2f})\nEarly completion!\n",
+                        'ja': f"✅ 品質満足 (スコア: {quality_score:.2f})\n早期完了!\n",
+                        'zh': f"✅ 质量满足 (分数: {quality_score:.2f})\n提前完成!\n"
+                    }
+                    agent_thoughts += quality_msg.get(lang, quality_msg['en'])
+                
+                yield accumulated_response, agent_thoughts
+                return
+            
+            # === 4단계: 최종 통합 (스트리밍) ===
+            if show_progress:
+                final_msg = {
+                    'ko': "✅ 최종 통합 중...\n\n",
+                    'en': "✅ Final integration in progress...\n\n",
+                    'ja': "✅ 最終統合中...\n\n",
+                    'zh': "✅ 最终整合中...\n\n"
+                }
+                agent_thoughts += final_msg.get(lang, final_msg['en'])
+                yield accumulated_response, agent_thoughts
+            
+            # 최종 프롬프트 (언어별)
+            final_prompt_templates = {
+                'ko': f"""
+질문: {query}
+창의성답변: {creative_response}
+비평피드백: {critic_response}
+감독자구조: {supervisor_response}
+최종통합→완벽답변. 마크다운 헤더(#, ##, ###) 사용 금지.""",
+                'en': f"""
+Question: {query}
+Creative answer: {creative_response}
+Critic feedback: {critic_response}
+Supervisor structure: {supervisor_response}
+Final integration→perfect answer. No markdown headers (#, ##, ###).""",
+                'ja': f"""
+質問: {query}
+創造的回答: {creative_response}
+批評フィードバック: {critic_response}
+監督者構造: {supervisor_response}
+最終統合→完璧な回答。マークダウンヘッダー（#、##、###）使用禁止。""",
+                'zh': f"""
+问题: {query}
+创意答案: {creative_response}
+评论反馈: {critic_response}
+主管结构: {supervisor_response}
+最终整合→完美答案。禁止使用Markdown标题（#、##、###）。"""
+            }
+            
+            final_prompt = final_prompt_templates.get(lang, final_prompt_templates['en'])
+            
+            final_task = self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.FINALIZER]},
+                    {"role": "user", "content": final_prompt}
+                ],
+                temperature=0.5,
+                max_tokens=2500
+            )
+            
+            # 최종 답변 스트리밍
+            accumulated_response = ""
+            
+            async for chunk in final_task:
+                accumulated_response += chunk
+                # 실시간 정리
+                cleaned_response = self.response_cleaner.clean_response(accumulated_response)
+                yield cleaned_response, agent_thoughts
+            
+            # 최종 정리
+            accumulated_response = self.response_cleaner.clean_response(accumulated_response)
+            
+            # 처리 시간 추가 (언어별)
+            processing_time = time.time() - start_time
+            time_msg = {
+                'ko': f"\n\n---\n⚡ 처리 시간: {processing_time:.1f}초",
+                'en': f"\n\n---\n⚡ Processing time: {processing_time:.1f} seconds",
+                'ja': f"\n\n---\n⚡ 処理時間: {processing_time:.1f}秒",
+                'zh': f"\n\n---\n⚡ 处理时间: {processing_time:.1f}秒"
+            }
+            accumulated_response += time_msg.get(lang, time_msg['en'])
+            
+            yield accumulated_response, agent_thoughts
+            
+        except Exception as e:
+            error_msg = {
+                'ko': f"❌ 오류 발생: {str(e)}",
+                'en': f"❌ Error occurred: {str(e)}",
+                'ja': f"❌ エラー発生: {str(e)}",
+                'zh': f"❌ 发生错误: {str(e)}"
+            }
+            yield error_msg.get(lang, error_msg['en']), agent_thoughts
+    
+    async def _run_critic_async(self, prompt: str) -> str:
+        """비평자 비동기 실행 with error handling"""
+        try:
+            response = ""
+            async for chunk in self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.CRITIC]},
+                    {"role": "user", "content": prompt}
+                ],
+                temperature=0.2,
+                max_tokens=500
+            ):
+                response += chunk
+            return response
+        except Exception as e:
+            # 언어 감지
+            lang = 'ko' if '질문' in prompt else 'en'
+            error_msg = {
+                'ko': "비평 처리 중 오류",
+                'en': "Error during critic processing",
+                'ja': "批評処理中のエラー",
+                'zh': "评论处理中出错"
+            }
+            return error_msg.get(lang, error_msg['en'])
+    
+    def _format_search_results(self, results: List[Dict]) -> str:
+        """검색 결과 압축 포맷"""
+        if not results:
+            return "No search results"
+        
+        formatted = []
+        for i, r in enumerate(results[:3], 1):
+            title = r.get('title', '')[:50]
+            desc = r.get('description', '')[:100]
+            formatted.append(f"[{i}]{title}:{desc}")
+        
+        return " | ".join(formatted)
+
+
+# ============================================================================
+# Gradio UI (최적화 버전 - 캐싱 제거)
+# ============================================================================
+
+def create_optimized_gradio_interface():
+    """최적화된 Gradio 인터페이스 (캐싱 없음)"""
+    
+    # 시스템 초기화
+    system = SpeedOptimizedMultiAgentSystem()
+    
+    def process_query_optimized(
+        message: str,
+        history: List[Dict],
+        use_search: bool,
+        show_agent_thoughts: bool,
+        search_count: int,
+        language_mode: str
+    ):
+        """최적화된 쿼리 처리 - 실시간 스트리밍 버전"""
+        
+        if not message:
+            yield history, "", ""
+            return
+        
+        # 언어 설정
+        if language_mode == "Auto":
+            lang = None  # 자동 감지
+        else:
+            lang_map = {"Korean": "ko", "English": "en", "Japanese": "ja", "Chinese": "zh"}
+            lang = lang_map.get(language_mode, None)
+        
+        # 비동기 함수를 동기적으로 실행
+        try:
+            import nest_asyncio
+            nest_asyncio.apply()
+        except ImportError:
+            pass
+        
+        try:
+            # 검색 수행 (동기화)
+            search_results = []
+            search_display = ""
+            
+            # 언어 자동 감지 (필요한 경우)
+            detected_lang = lang or system.language_detector.detect_language(message)
+            
+            if use_search:
+                # 검색 상태 표시
+                processing_msg = {
+                    'ko': "⚡ 고속 처리 중...",
+                    'en': "⚡ High-speed processing...",
+                    'ja': "⚡ 高速処理中...",
+                    'zh': "⚡ 高速处理中..."
+                }
+                history_with_message = history + [
+                    {"role": "user", "content": message},
+                    {"role": "assistant", "content": processing_msg.get(detected_lang, processing_msg['en'])}
+                ]
+                yield history_with_message, "", ""
+                
+                # 비동기 검색을 동기적으로 실행
+                async def search_wrapper():
+                    return await system.search.search_async(message, count=search_count, lang=detected_lang)
+                
+                loop = asyncio.new_event_loop()
+                asyncio.set_event_loop(loop)
+                search_results = loop.run_until_complete(search_wrapper())
+                
+                if search_results:
+                    ref_label = {
+                        'ko': "📚 참고 자료",
+                        'en': "📚 References",
+                        'ja': "📚 参考資料",
+                        'zh': "📚 参考资料"
+                    }
+                    search_display = f"{ref_label.get(detected_lang, ref_label['en'])}\n\n"
+                    for i, result in enumerate(search_results[:3], 1):
+                        search_display += f"**{i}. [{result['title'][:50]}]({result['url']})**\n"
+                        search_display += f"   {result['description'][:100]}...\n\n"
+            
+            # 사용자 메시지 추가
+            current_history = history + [{"role": "user", "content": message}]
+            
+            # 실시간 스트리밍을 위한 비동기 처리
+            async def stream_responses():
+                """실시간 스트리밍 제너레이터"""
+                async for response, thoughts in system.parallel_process_agents(
+                    query=message,
+                    search_results=search_results,
+                    show_progress=show_agent_thoughts,
+                    lang=detected_lang
+                ):
+                    yield response, thoughts
+            
+            # 새 이벤트 루프에서 실시간 스트리밍
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+            
+            # 비동기 제너레이터를 동기적으로 순회
+            gen = stream_responses()
+            
+            while True:
+                try:
+                    # 다음 항목 가져오기
+                    task = asyncio.ensure_future(gen.__anext__(), loop=loop)
+                    response, thoughts = loop.run_until_complete(task)
+                    
+                    # 실시간 업데이트
+                    updated_history = current_history + [
+                        {"role": "assistant", "content": response}
+                    ]
+                    yield updated_history, thoughts, search_display
+                    
+                except StopAsyncIteration:
+                    break
+            
+        except Exception as e:
+            error_history = history + [
+                {"role": "user", "content": message},
+                {"role": "assistant", "content": f"❌ Error: {str(e)}"}
+            ]
+            yield error_history, "", ""
+        finally:
+            # 루프 정리
+            try:
+                loop.close()
+            except:
+                pass
+    
+    # Gradio 인터페이스
+    with gr.Blocks(
+        title="⚡ Speed-Optimized Multi-Agent System (No Cache)",
+        theme=gr.themes.Soft(),
+        css="""
+        .gradio-container {
+            max-width: 1400px !important;
+            margin: auto !important;
+        }
+        """
+    ) as demo:
+        gr.Markdown("""
+        # ⚡ Enhanced Multi-Agent RAG System (캐싱 제거 버전)
+        **Complex questions processed within 5-8 seconds | Multi-language support**
+        
+        **Optimization Features:**
+        - 🚀 Parallel Processing: Concurrent agent execution
+        - ⚡ Stream Buffering: Network optimization
+        - 🎯 Early Termination: Complete immediately when quality is met
+        - 🌍 Multi-language: Auto-detect Korean/English/Japanese/Chinese
+        - ❌ **Caching Disabled**: 캐싱 기능 제거됨
+        """)
+        
+        with gr.Row():
+            with gr.Column(scale=3):
+                chatbot = gr.Chatbot(
+                    height=500,
+                    label="💬 Chat",
+                    type="messages"
+                )
+                
+                msg = gr.Textbox(
+                    label="Enter complex question",
+                    placeholder="Enter complex questions requiring analysis, strategy, or creative solutions...",
+                    lines=3
+                )
+                
+                with gr.Row():
+                    submit = gr.Button("⚡ High-Speed Process", variant="primary")
+                    clear = gr.Button("🔄 Reset")
+                
+                with gr.Accordion("🤖 Agent Processing", open=False):
+                    agent_thoughts = gr.Markdown()
+                
+                with gr.Accordion("📚 Search Sources", open=False):
+                    search_sources = gr.Markdown()
+            
+            with gr.Column(scale=1):
+                gr.Markdown("**⚙️ Settings**")
+                
+                language_mode = gr.Radio(
+                    choices=["Auto", "Korean", "English", "Japanese", "Chinese"],
+                    value="Auto",
+                    label="🌍 Language Mode"
+                )
+                
+                use_search = gr.Checkbox(
+                    label="🔍 Use Web Search",
+                    value=True
+                )
+                
+                show_agent_thoughts = gr.Checkbox(
+                    label="🧠 Show Processing",
+                    value=True
+                )
+                
+                search_count = gr.Slider(
+                    minimum=3,
+                    maximum=10,
+                    value=5,
+                    step=1,
+                    label="Search Results Count"
+                )
+                
+                gr.Markdown("""
+                **⚡ Optimization Status**
+                
+                **Active Optimizations:**
+                - ✅ Parallel Processing
+                - ❌ ~~Smart Caching~~ (제거됨)
+                - ✅ Buffer Streaming
+                - ✅ Early Termination
+                - ✅ Compressed Prompts
+                - ✅ Multi-language Support
+                - ✅ Error Recovery
+                
+                **Expected Processing Time:**
+                - Simple Query: 3-5 seconds
+                - Complex Query: 5-8 seconds
+                - Very Complex: 8-12 seconds
+                """)
+        
+        # 복잡한 질문 예제 (다국어)
+        gr.Examples(
+            examples=[
+                # Korean
+                "AI 기술이 향후 10년간 한국 경제에 미칠 영향을 다각도로 분석하고 대응 전략을 제시해줘",
+                "스타트업이 대기업과 경쟁하기 위한 혁신적인 전략을 단계별로 수립해줘",
+                # English
+                "Analyze the multifaceted impact of quantum computing on current encryption systems and propose alternatives",
+                "Design 5 innovative business models for climate change mitigation with practical implementation details",
+                # Japanese
+                "メタバース時代の教育革新方案を実装可能なレベルで提案してください",
+                # Chinese
+                "分析人工智能对未来十年全球经济的影响并提出应对策略"
+            ],
+            inputs=msg
+        )
+        
+        # 이벤트 바인딩
+        submit.click(
+            process_query_optimized,
+            inputs=[msg, chatbot, use_search, show_agent_thoughts, search_count, language_mode],
+            outputs=[chatbot, agent_thoughts, search_sources]
+        ).then(
+            lambda: "",
+            None,
+            msg
+        )
+        
+        msg.submit(
+            process_query_optimized,
+            inputs=[msg, chatbot, use_search, show_agent_thoughts, search_count, language_mode],
+            outputs=[chatbot, agent_thoughts, search_sources]
+        ).then(
+            lambda: "",
+            None,
+            msg
+        )
+        
+        clear.click(
+            lambda: ([], "", ""),
+            None,
+            [chatbot, agent_thoughts, search_sources]
+        )
+    
+    return demo
+
+
+# ============================================================================
+# 메인 실행
+# ============================================================================
+
+if __name__ == "__main__":
+    print("""
+╔══════════════════════════════════════════════════════════════╗
+║     ⚡ Speed-Optimized Multi-Agent System (No Cache) ⚡     ║
+║                                                              ║
+║   High-speed AI system processing complex questions         ║
+║                                                              ║
+║  Features:                                                  ║
+║  • Multi-language support (KO/EN/JA/ZH)                    ║
+║  • Improved error recovery                                  ║
+║  • NO CACHING (캐싱 기능 제거됨)                           ║
+║  • Adaptive stream buffering                                ║
+║  • Response cleaning & formatting                           ║
+╚══════════════════════════════════════════════════════════════╝
+    """)
+    
+    # API 키 확인
+    if not os.getenv("FIREWORKS_API_KEY"):
+        print("\n⚠️  FIREWORKS_API_KEY is not set.")
+    
+    if not os.getenv("BRAVE_SEARCH_API_KEY"):
+        print("\n⚠️  BRAVE_SEARCH_API_KEY is not set.")
+    
+    # Gradio 앱 실행
+    demo = create_optimized_gradio_interface()
+    
+    is_hf_spaces = os.getenv("SPACE_ID") is not None
+    
+    if is_hf_spaces:
+        print("\n🤗 Running in optimized mode on Hugging Face Spaces (No Cache)...")
+        demo.launch(server_name="0.0.0.0", server_port=7860)
+    else:
+        print("\n💻 Running in optimized mode on local environment (No Cache)...")
+        demo.launch(server_name="0.0.0.0", server_port=7860, share=False), '', response, flags=re.MULTILINE)
+        
+        # 테이블 구분선 제거 (|---|, |:---|, |---:|, |:---:| 등)
+        response = re.sub(r'^\|[\s\-:]+\|.*
+
+
+# ============================================================================
+# 통합 최적화 멀티 에이전트 시스템 (캐싱 제거 버전)
+# ============================================================================
+
+class SpeedOptimizedMultiAgentSystem:
+    """속도 최적화된 멀티 에이전트 시스템 (캐싱 없음)"""
+    
+    def __init__(self):
+        self.llm = OptimizedFireworksClient()
+        self.search = AsyncBraveSearch()
+        self.reasoning = LightweightReasoningChain()
+        self.quality_checker = QualityChecker()
+        self.streaming = OptimizedStreaming()
+        self.language_detector = LanguageDetector()
+        self.response_cleaner = ResponseCleaner()
+        
+        # 병렬 처리 풀
+        self.executor = ThreadPoolExecutor(max_workers=4)
+    
+    def _init_compact_prompts(self, lang: str = 'ko') -> Dict:
+        """압축된 고효율 프롬프트 (언어별)"""
+        prompts = {
+            'ko': {
+                AgentRole.SUPERVISOR: """[감독자-구조설계]
+즉시분석: 핵심의도+필요정보+답변구조
+출력: 5개 핵심포인트(각 1문장)
+추론체계 명시""",
+                
+                AgentRole.CREATIVE: """[창의성생성자]
+입력구조 따라 창의적 확장
+실용예시+혁신접근+구체조언
+불필요설명 제거""",
+                
+                AgentRole.CRITIC: """[비평자-검증]
+신속검토: 정확성/논리성/실용성
+개선포인트 3개만
+각 2문장 이내""",
+                
+                AgentRole.FINALIZER: """[최종통합]
+모든의견 종합→최적답변
+명확구조+실용정보+창의균형
+바로 핵심 내용부터 시작. 불필요한 헤더나 마크업 없이. 마크다운 헤더(#, ##, ###) 사용 금지."""
+            },
+            'en': {
+                AgentRole.SUPERVISOR: """[Supervisor-Structure]
+Immediate analysis: core intent+required info+answer structure
+Output: 5 key points (1 sentence each)
+Clear reasoning framework""",
+                
+                AgentRole.CREATIVE: """[Creative Generator]
+Follow structure, expand creatively
+Practical examples+innovative approach+specific advice
+Remove unnecessary explanations""",
+                
+                AgentRole.CRITIC: """[Critic-Verification]
+Quick review: accuracy/logic/practicality
+Only 3 improvement points
+Max 2 sentences each""",
+                
+                AgentRole.FINALIZER: """[Final Integration]
+Synthesize all inputs→optimal answer
+Clear structure+practical info+creative balance
+Start with core content directly. No unnecessary headers or markup. No markdown headers (#, ##, ###)."""
+            },
+            'ja': {
+                AgentRole.SUPERVISOR: """[監督者-構造設計]
+即時分析：核心意図+必要情報+回答構造
+出力：5つの核心ポイント（各1文）
+推論体系明示""",
+                
+                AgentRole.CREATIVE: """[創造性生成者]
+入力構造に従って創造的拡張
+実用例+革新的アプローチ+具体的アドバイス
+不要な説明削除""",
+                
+                AgentRole.CRITIC: """[批評者-検証]
+迅速レビュー：正確性/論理性/実用性
+改善ポイント3つのみ
+各2文以内""",
+                
+                AgentRole.FINALIZER: """[最終統合]
+全意見統合→最適回答
+明確構造+実用情報+創造性���ランス
+核心内容から直接開始。不要なヘッダーやマークアップなし。マークダウンヘッダー（#、##、###）使用禁止。"""
+            },
+            'zh': {
+                AgentRole.SUPERVISOR: """[主管-结构设计]
+立即分析：核心意图+所需信息+答案结构
+输出：5个核心要点（每个1句）
+推理体系明确""",
+                
+                AgentRole.CREATIVE: """[创意生成器]
+按结构创造性扩展
+实用示例+创新方法+具体建议
+删除不必要的解释""",
+                
+                AgentRole.CRITIC: """[评论家-验证]
+快速审查：准确性/逻辑性/实用性
+仅3个改进点
+每个最多2句""",
+                
+                AgentRole.FINALIZER: """[最终整合]
+综合所有意见→最佳答案
+清晰结构+实用信息+创意平衡
+直接从核心内容开始。无需不必要的标题或标记。禁止使用Markdown标题（#、##、###）。"""
+            }
+        }
+        
+        return prompts.get(lang, prompts['en'])
+    
+    async def parallel_process_agents(
+        self,
+        query: str,
+        search_results: List[Dict],
+        show_progress: bool = True,
+        lang: str = None
+    ) -> AsyncGenerator[Tuple[str, str], None]:
+        """병렬 처리 파이프라인 (캐싱 없음)"""
+        
+        start_time = time.time()
+        
+        # 언어 자동 감지
+        if lang is None:
+            lang = self.language_detector.detect_language(query)
+        
+        # 언어별 프롬프트 설정
+        self.compact_prompts = self._init_compact_prompts(lang)
+        
+        search_context = self._format_search_results(search_results)
+        accumulated_response = ""
+        agent_thoughts = ""
+        
+        # 추론 패턴 결정
+        reasoning_pattern = self.reasoning.get_reasoning_pattern(query, lang)
+        
+        try:
+            # === 1단계: 감독자 + 검색 병렬 실행 ===
+            if show_progress:
+                progress_msg = {
+                    'ko': "🚀 병렬 처리 시작\n👔 감독자 분석 + 🔍 추가 검색 동시 진행...\n\n",
+                    'en': "🚀 Starting parallel processing\n👔 Supervisor analysis + 🔍 Additional search in progress...\n\n",
+                    'ja': "🚀 並列処理開始\n👔 監督者分析 + 🔍 追加検索同時進行中...\n\n",
+                    'zh': "🚀 开始并行处理\n👔 主管分析 + 🔍 附加搜索同时进行...\n\n"
+                }
+                agent_thoughts = progress_msg.get(lang, progress_msg['en'])
+                yield accumulated_response, agent_thoughts
+            
+            # 감독자 프롬프트 (언어별)
+            supervisor_prompt_templates = {
+                'ko': f"""
+질문: {query}
+검색결과: {search_context}
+추론패턴: {reasoning_pattern}
+즉시 핵심구조 5개 제시""",
+                'en': f"""
+Question: {query}
+Search results: {search_context}
+Reasoning pattern: {reasoning_pattern}
+Immediately provide 5 key structures""",
+                'ja': f"""
+質問: {query}
+検索結果: {search_context}
+推論パターン: {reasoning_pattern}
+即座に5つの核心構造を提示""",
+                'zh': f"""
+问题: {query}
+搜索结果: {search_context}
+推理模式: {reasoning_pattern}
+立即提供5个核心结构"""
+            }
+            
+            supervisor_prompt = supervisor_prompt_templates.get(lang, supervisor_prompt_templates['en'])
+            
+            supervisor_response = ""
+            supervisor_task = self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.SUPERVISOR]},
+                    {"role": "user", "content": supervisor_prompt}
+                ],
+                temperature=0.3,
+                max_tokens=500
+            )
+            
+            # 감독자 스트리밍 (버퍼링)
+            async for chunk in self.streaming.buffer_and_yield(supervisor_task):
+                supervisor_response += chunk
+                if show_progress and len(supervisor_response) < 300:
+                    supervisor_label = {
+                        'ko': "👔 감독자 분석",
+                        'en': "👔 Supervisor Analysis",
+                        'ja': "👔 監督者分析",
+                        'zh': "👔 主管分析"
+                    }
+                    agent_thoughts = f"{supervisor_label.get(lang, supervisor_label['en'])}\n{supervisor_response[:300]}...\n\n"
+                    yield accumulated_response, agent_thoughts
+            
+            # === 2단계: 창의성 + 비평 준비 병렬 ===
+            if show_progress:
+                creative_msg = {
+                    'ko': "🎨 창의성 생성자 + 🔍 비평자 준비...\n\n",
+                    'en': "🎨 Creative Generator + 🔍 Critic preparing...\n\n",
+                    'ja': "🎨 創造性生成者 + 🔍 批評者準備中...\n\n",
+                    'zh': "🎨 创意生成器 + 🔍 评论家准备中...\n\n"
+                }
+                agent_thoughts += creative_msg.get(lang, creative_msg['en'])
+                yield accumulated_response, agent_thoughts
+            
+            # 창의성 생성 시작 (언어별)
+            creative_prompt_templates = {
+                'ko': f"""
+질문: {query}
+감독자구조: {supervisor_response}
+검색결과: {search_context}
+창의적+실용적 답변 즉시생성""",
+                'en': f"""
+Question: {query}
+Supervisor structure: {supervisor_response}
+Search results: {search_context}
+Generate creative+practical answer immediately""",
+                'ja': f"""
+質問: {query}
+監督者構造: {supervisor_response}
+検索結果: {search_context}
+創造的+実用的回答即座生成""",
+                'zh': f"""
+问题: {query}
+主管结构: {supervisor_response}
+搜索结果: {search_context}
+立即生成创意+实用答案"""
+            }
+            
+            creative_prompt = creative_prompt_templates.get(lang, creative_prompt_templates['en'])
+            
+            creative_response = ""
+            creative_partial = ""
+            critic_started = False
+            critic_response = ""
+            
+            creative_task = self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.CREATIVE]},
+                    {"role": "user", "content": creative_prompt}
+                ],
+                temperature=0.8,
+                max_tokens=1500
+            )
+            
+            # 창의성 스트리밍 + 비평자 조기 시작
+            async for chunk in self.streaming.buffer_and_yield(creative_task):
+                creative_response += chunk
+                creative_partial += chunk
+                
+                # 창의성 응답이 500자 넘으면 비평자 시작
+                if len(creative_partial) > 500 and not critic_started:
+                    critic_started = True
+                    
+                    # 비평자 비동기 시작 (언어별)
+                    critic_prompt_templates = {
+                        'ko': f"""
+원본질문: {query}
+창의성답변(일부): {creative_partial}
+신속검토→개선점3개""",
+                        'en': f"""
+Original question: {query}
+Creative answer (partial): {creative_partial}
+Quick review→3 improvements""",
+                        'ja': f"""
+元の質問: {query}
+創造的回答（一部）: {creative_partial}
+迅速レビュー→改善点3つ""",
+                        'zh': f"""
+原始问题: {query}
+创意答案（部分）: {creative_partial}
+快速审查→3个改进点"""
+                    }
+                    
+                    critic_prompt = critic_prompt_templates.get(lang, critic_prompt_templates['en'])
+                    
+                    critic_task = asyncio.create_task(
+                        self._run_critic_async(critic_prompt)
+                    )
+                
+                if show_progress:
+                    display_creative = creative_response[:400] + "..." if len(creative_response) > 400 else creative_response
+                    creative_label = {
+                        'ko': "🎨 창의성 생성자",
+                        'en': "🎨 Creative Generator",
+                        'ja': "🎨 創造性生成者",
+                        'zh': "🎨 创意生成器"
+                    }
+                    agent_thoughts = f"{creative_label.get(lang, creative_label['en'])}\n{display_creative}\n\n"
+                    yield accumulated_response, agent_thoughts
+            
+            # 비평자 결과 대기
+            if critic_started:
+                critic_response = await critic_task
+                
+                if show_progress:
+                    critic_label = {
+                        'ko': "🔍 비평자 검토",
+                        'en': "🔍 Critic Review",
+                        'ja': "🔍 批評者レビュー",
+                        'zh': "🔍 评论家审查"
+                    }
+                    agent_thoughts += f"{critic_label.get(lang, critic_label['en'])}\n{critic_response[:200]}...\n\n"
+                    yield accumulated_response, agent_thoughts
+            
+            # === 3단계: 품질 체크 및 조기 종료 ===
+            quality_score, need_more = self.quality_checker.evaluate_response(
+                creative_response, query, lang
+            )
+            
+            if not need_more and quality_score > 0.85:
+                # 품질이 충분히 높으면 바로 반환
+                accumulated_response = self.response_cleaner.clean_response(creative_response)
+                
+                if show_progress:
+                    quality_msg = {
+                        'ko': f"✅ 품질 충족 (점수: {quality_score:.2f})\n조기 완료!\n",
+                        'en': f"✅ Quality met (score: {quality_score:.2f})\nEarly completion!\n",
+                        'ja': f"✅ 品質満足 (スコア: {quality_score:.2f})\n早期完了!\n",
+                        'zh': f"✅ 质量满足 (分数: {quality_score:.2f})\n提前完成!\n"
+                    }
+                    agent_thoughts += quality_msg.get(lang, quality_msg['en'])
+                
+                yield accumulated_response, agent_thoughts
+                return
+            
+            # === 4단계: 최종 통합 (스트리밍) ===
+            if show_progress:
+                final_msg = {
+                    'ko': "✅ 최종 통합 중...\n\n",
+                    'en': "✅ Final integration in progress...\n\n",
+                    'ja': "✅ 最終統合中...\n\n",
+                    'zh': "✅ 最终整合中...\n\n"
+                }
+                agent_thoughts += final_msg.get(lang, final_msg['en'])
+                yield accumulated_response, agent_thoughts
+            
+            # 최종 프롬프트 (언어별)
+            final_prompt_templates = {
+                'ko': f"""
+질문: {query}
+창의성답변: {creative_response}
+비평피드백: {critic_response}
+감독자구조: {supervisor_response}
+최종통합→완벽답변. 마크다운 헤더(#, ##, ###) 사용 금지.""",
+                'en': f"""
+Question: {query}
+Creative answer: {creative_response}
+Critic feedback: {critic_response}
+Supervisor structure: {supervisor_response}
+Final integration→perfect answer. No markdown headers (#, ##, ###).""",
+                'ja': f"""
+質問: {query}
+創造的回答: {creative_response}
+批評フィードバック: {critic_response}
+監督者構造: {supervisor_response}
+最終統合→完璧な回答。マークダウンヘッダー（#、##、###）使用禁止。""",
+                'zh': f"""
+问题: {query}
+创意答案: {creative_response}
+评论反馈: {critic_response}
+主管结构: {supervisor_response}
+最终整合→完美答案。禁止使用Markdown标题（#、##、###）。"""
+            }
+            
+            final_prompt = final_prompt_templates.get(lang, final_prompt_templates['en'])
+            
+            final_task = self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.FINALIZER]},
+                    {"role": "user", "content": final_prompt}
+                ],
+                temperature=0.5,
+                max_tokens=2500
+            )
+            
+            # 최종 답변 스트리밍
+            accumulated_response = ""
+            
+            async for chunk in final_task:
+                accumulated_response += chunk
+                # 실시간 정리
+                cleaned_response = self.response_cleaner.clean_response(accumulated_response)
+                yield cleaned_response, agent_thoughts
+            
+            # 최종 정리
+            accumulated_response = self.response_cleaner.clean_response(accumulated_response)
+            
+            # 처리 시간 추가 (언어별)
+            processing_time = time.time() - start_time
+            time_msg = {
+                'ko': f"\n\n---\n⚡ 처리 시간: {processing_time:.1f}초",
+                'en': f"\n\n---\n⚡ Processing time: {processing_time:.1f} seconds",
+                'ja': f"\n\n---\n⚡ 処理時間: {processing_time:.1f}秒",
+                'zh': f"\n\n---\n⚡ 处理时间: {processing_time:.1f}秒"
+            }
+            accumulated_response += time_msg.get(lang, time_msg['en'])
+            
+            yield accumulated_response, agent_thoughts
+            
+        except Exception as e:
+            error_msg = {
+                'ko': f"❌ 오류 발생: {str(e)}",
+                'en': f"❌ Error occurred: {str(e)}",
+                'ja': f"❌ エラー発生: {str(e)}",
+                'zh': f"❌ 发生错误: {str(e)}"
+            }
+            yield error_msg.get(lang, error_msg['en']), agent_thoughts
+    
+    async def _run_critic_async(self, prompt: str) -> str:
+        """비평자 비동기 실행 with error handling"""
+        try:
+            response = ""
+            async for chunk in self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.CRITIC]},
+                    {"role": "user", "content": prompt}
+                ],
+                temperature=0.2,
+                max_tokens=500
+            ):
+                response += chunk
+            return response
+        except Exception as e:
+            # 언어 감지
+            lang = 'ko' if '질문' in prompt else 'en'
+            error_msg = {
+                'ko': "비평 처리 중 오류",
+                'en': "Error during critic processing",
+                'ja': "批評処理中のエラー",
+                'zh': "评论处理中出错"
+            }
+            return error_msg.get(lang, error_msg['en'])
+    
+    def _format_search_results(self, results: List[Dict]) -> str:
+        """검색 결과 압축 포맷"""
+        if not results:
+            return "No search results"
+        
+        formatted = []
+        for i, r in enumerate(results[:3], 1):
+            title = r.get('title', '')[:50]
+            desc = r.get('description', '')[:100]
+            formatted.append(f"[{i}]{title}:{desc}")
+        
+        return " | ".join(formatted)
+
+
+# ============================================================================
+# Gradio UI (최적화 버전 - 캐싱 제거)
+# ============================================================================
+
+def create_optimized_gradio_interface():
+    """최적화된 Gradio 인터페이스 (캐싱 없음)"""
+    
+    # 시스템 초기화
+    system = SpeedOptimizedMultiAgentSystem()
+    
+    def process_query_optimized(
+        message: str,
+        history: List[Dict],
+        use_search: bool,
+        show_agent_thoughts: bool,
+        search_count: int,
+        language_mode: str
+    ):
+        """최적화된 쿼리 처리 - 실시간 스트리밍 버전"""
+        
+        if not message:
+            yield history, "", ""
+            return
+        
+        # 언어 설정
+        if language_mode == "Auto":
+            lang = None  # 자동 감지
+        else:
+            lang_map = {"Korean": "ko", "English": "en", "Japanese": "ja", "Chinese": "zh"}
+            lang = lang_map.get(language_mode, None)
+        
+        # 비동기 함수를 동기적으로 실행
+        try:
+            import nest_asyncio
+            nest_asyncio.apply()
+        except ImportError:
+            pass
+        
+        try:
+            # 검색 수행 (동기화)
+            search_results = []
+            search_display = ""
+            
+            # 언어 자동 감지 (필요한 경우)
+            detected_lang = lang or system.language_detector.detect_language(message)
+            
+            if use_search:
+                # 검색 상태 표시
+                processing_msg = {
+                    'ko': "⚡ 고속 처리 중...",
+                    'en': "⚡ High-speed processing...",
+                    'ja': "⚡ 高速処理中...",
+                    'zh': "⚡ 高速处理中..."
+                }
+                history_with_message = history + [
+                    {"role": "user", "content": message},
+                    {"role": "assistant", "content": processing_msg.get(detected_lang, processing_msg['en'])}
+                ]
+                yield history_with_message, "", ""
+                
+                # 비동기 검색을 동기적으로 실행
+                async def search_wrapper():
+                    return await system.search.search_async(message, count=search_count, lang=detected_lang)
+                
+                loop = asyncio.new_event_loop()
+                asyncio.set_event_loop(loop)
+                search_results = loop.run_until_complete(search_wrapper())
+                
+                if search_results:
+                    ref_label = {
+                        'ko': "📚 참고 자료",
+                        'en': "📚 References",
+                        'ja': "📚 参考資料",
+                        'zh': "📚 参考资料"
+                    }
+                    search_display = f"{ref_label.get(detected_lang, ref_label['en'])}\n\n"
+                    for i, result in enumerate(search_results[:3], 1):
+                        search_display += f"**{i}. [{result['title'][:50]}]({result['url']})**\n"
+                        search_display += f"   {result['description'][:100]}...\n\n"
+            
+            # 사용자 메시지 추가
+            current_history = history + [{"role": "user", "content": message}]
+            
+            # 실시간 스트리밍을 위한 비동기 처리
+            async def stream_responses():
+                """실시간 스트리밍 제너레이터"""
+                async for response, thoughts in system.parallel_process_agents(
+                    query=message,
+                    search_results=search_results,
+                    show_progress=show_agent_thoughts,
+                    lang=detected_lang
+                ):
+                    yield response, thoughts
+            
+            # 새 이벤트 루프에서 실시간 스트리밍
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+            
+            # 비동기 제너레이터를 동기적으로 순회
+            gen = stream_responses()
+            
+            while True:
+                try:
+                    # 다음 항목 가져오기
+                    task = asyncio.ensure_future(gen.__anext__(), loop=loop)
+                    response, thoughts = loop.run_until_complete(task)
+                    
+                    # 실시간 업데이트
+                    updated_history = current_history + [
+                        {"role": "assistant", "content": response}
+                    ]
+                    yield updated_history, thoughts, search_display
+                    
+                except StopAsyncIteration:
+                    break
+            
+        except Exception as e:
+            error_history = history + [
+                {"role": "user", "content": message},
+                {"role": "assistant", "content": f"❌ Error: {str(e)}"}
+            ]
+            yield error_history, "", ""
+        finally:
+            # 루프 정리
+            try:
+                loop.close()
+            except:
+                pass
+    
+    # Gradio 인터페이스
+    with gr.Blocks(
+        title="⚡ Speed-Optimized Multi-Agent System (No Cache)",
+        theme=gr.themes.Soft(),
+        css="""
+        .gradio-container {
+            max-width: 1400px !important;
+            margin: auto !important;
+        }
+        """
+    ) as demo:
+        gr.Markdown("""
+        # ⚡ Enhanced Multi-Agent RAG System (캐싱 제거 버전)
+        **Complex questions processed within 5-8 seconds | Multi-language support**
+        
+        **Optimization Features:**
+        - 🚀 Parallel Processing: Concurrent agent execution
+        - ⚡ Stream Buffering: Network optimization
+        - 🎯 Early Termination: Complete immediately when quality is met
+        - 🌍 Multi-language: Auto-detect Korean/English/Japanese/Chinese
+        - ❌ **Caching Disabled**: 캐싱 기능 제거됨
+        """)
+        
+        with gr.Row():
+            with gr.Column(scale=3):
+                chatbot = gr.Chatbot(
+                    height=500,
+                    label="💬 Chat",
+                    type="messages"
+                )
+                
+                msg = gr.Textbox(
+                    label="Enter complex question",
+                    placeholder="Enter complex questions requiring analysis, strategy, or creative solutions...",
+                    lines=3
+                )
+                
+                with gr.Row():
+                    submit = gr.Button("⚡ High-Speed Process", variant="primary")
+                    clear = gr.Button("🔄 Reset")
+                
+                with gr.Accordion("🤖 Agent Processing", open=False):
+                    agent_thoughts = gr.Markdown()
+                
+                with gr.Accordion("📚 Search Sources", open=False):
+                    search_sources = gr.Markdown()
+            
+            with gr.Column(scale=1):
+                gr.Markdown("**⚙️ Settings**")
+                
+                language_mode = gr.Radio(
+                    choices=["Auto", "Korean", "English", "Japanese", "Chinese"],
+                    value="Auto",
+                    label="🌍 Language Mode"
+                )
+                
+                use_search = gr.Checkbox(
+                    label="🔍 Use Web Search",
+                    value=True
+                )
+                
+                show_agent_thoughts = gr.Checkbox(
+                    label="🧠 Show Processing",
+                    value=True
+                )
+                
+                search_count = gr.Slider(
+                    minimum=3,
+                    maximum=10,
+                    value=5,
+                    step=1,
+                    label="Search Results Count"
+                )
+                
+                gr.Markdown("""
+                **⚡ Optimization Status**
+                
+                **Active Optimizations:**
+                - ✅ Parallel Processing
+                - ❌ ~~Smart Caching~~ (제거됨)
+                - ✅ Buffer Streaming
+                - ✅ Early Termination
+                - ✅ Compressed Prompts
+                - ✅ Multi-language Support
+                - ✅ Error Recovery
+                
+                **Expected Processing Time:**
+                - Simple Query: 3-5 seconds
+                - Complex Query: 5-8 seconds
+                - Very Complex: 8-12 seconds
+                """)
+        
+        # 복잡한 질문 예제 (다국어)
+        gr.Examples(
+            examples=[
+                # Korean
+                "AI 기술이 향후 10년간 한국 경제에 미칠 영향을 다각도로 분석하고 대응 전략을 제시해줘",
+                "스타트업이 대기업과 경쟁하기 위한 혁신적인 전략을 단계별로 수립해줘",
+                # English
+                "Analyze the multifaceted impact of quantum computing on current encryption systems and propose alternatives",
+                "Design 5 innovative business models for climate change mitigation with practical implementation details",
+                # Japanese
+                "メタバース時代の教育革新方案を実装可能なレベルで提案してください",
+                # Chinese
+                "分析人工智能对未来十年全球经济的影响并提出应对策略"
+            ],
+            inputs=msg
+        )
+        
+        # 이벤트 바인딩
+        submit.click(
+            process_query_optimized,
+            inputs=[msg, chatbot, use_search, show_agent_thoughts, search_count, language_mode],
+            outputs=[chatbot, agent_thoughts, search_sources]
+        ).then(
+            lambda: "",
+            None,
+            msg
+        )
+        
+        msg.submit(
+            process_query_optimized,
+            inputs=[msg, chatbot, use_search, show_agent_thoughts, search_count, language_mode],
+            outputs=[chatbot, agent_thoughts, search_sources]
+        ).then(
+            lambda: "",
+            None,
+            msg
+        )
+        
+        clear.click(
+            lambda: ([], "", ""),
+            None,
+            [chatbot, agent_thoughts, search_sources]
+        )
+    
+    return demo
+
+
+# ============================================================================
+# 메인 실행
+# ============================================================================
+
+if __name__ == "__main__":
+    print("""
+╔══════════════════════════════════════════════════════════════╗
+║     ⚡ Speed-Optimized Multi-Agent System (No Cache) ⚡     ║
+║                                                              ║
+║   High-speed AI system processing complex questions         ║
+║                                                              ║
+║  Features:                                                  ║
+║  • Multi-language support (KO/EN/JA/ZH)                    ║
+║  • Improved error recovery                                  ║
+║  • NO CACHING (캐싱 기능 제거됨)                           ║
+║  • Adaptive stream buffering                                ║
+║  • Response cleaning & formatting                           ║
+╚══════════════════════════════════════════════════════════════╝
+    """)
+    
+    # API 키 확인
+    if not os.getenv("FIREWORKS_API_KEY"):
+        print("\n⚠️  FIREWORKS_API_KEY is not set.")
+    
+    if not os.getenv("BRAVE_SEARCH_API_KEY"):
+        print("\n⚠️  BRAVE_SEARCH_API_KEY is not set.")
+    
+    # Gradio 앱 실행
+    demo = create_optimized_gradio_interface()
+    
+    is_hf_spaces = os.getenv("SPACE_ID") is not None
+    
+    if is_hf_spaces:
+        print("\n🤗 Running in optimized mode on Hugging Face Spaces (No Cache)...")
+        demo.launch(server_name="0.0.0.0", server_port=7860)
+    else:
+        print("\n💻 Running in optimized mode on local environment (No Cache)...")
+        demo.launch(server_name="0.0.0.0", server_port=7860, share=False), '', response, flags=re.MULTILINE)
+        
         # 마크다운 헤더 제거
         response = re.sub(r'^#{1,6}\s+', '', response, flags=re.MULTILINE)
         
-        # 불필요한 구분선 제거
-        response = re.sub(r'\*{2,}|_{2,}|-{3,}', '', response)
+        # 볼드 텍스트 제거 (**text** 또는 __text__)
+        response = re.sub(r'\*\*(.*?)\*\*', r'\1', response)
+        response = re.sub(r'__(.*?)__', r'\1', response)
+        
+        # 이탤릭 텍스트 제거 (*text* 또는 _text_)
+        response = re.sub(r'\*(.*?)\*', r'\1', response)
+        response = re.sub(r'_(.*?)_', r'\1', response)
+        
+        # 코드 블록 제거 (```code```)
+        response = re.sub(r'```[\s\S]*?```', '', response)
+        
+        # 인라인 코드 제거 (`code`)
+        response = re.sub(r'`([^`]*)`', r'\1', response)
+        
+        # 불필요한 구분선 제거 (---, ***, ___)
+        response = re.sub(r'^[\*\-_]{3,}
+
+
+# ============================================================================
+# 통합 최적화 멀티 에이전트 시스템 (캐싱 제거 버전)
+# ============================================================================
+
+class SpeedOptimizedMultiAgentSystem:
+    """속도 최적화된 멀티 에이전트 시스템 (캐싱 없음)"""
+    
+    def __init__(self):
+        self.llm = OptimizedFireworksClient()
+        self.search = AsyncBraveSearch()
+        self.reasoning = LightweightReasoningChain()
+        self.quality_checker = QualityChecker()
+        self.streaming = OptimizedStreaming()
+        self.language_detector = LanguageDetector()
+        self.response_cleaner = ResponseCleaner()
+        
+        # 병렬 처리 풀
+        self.executor = ThreadPoolExecutor(max_workers=4)
+    
+    def _init_compact_prompts(self, lang: str = 'ko') -> Dict:
+        """압축된 고효율 프롬프트 (언어별)"""
+        prompts = {
+            'ko': {
+                AgentRole.SUPERVISOR: """[감독자-구조설계]
+즉시분석: 핵심의도+필요정보+답변구조
+출력: 5개 핵심포인트(각 1문장)
+추론체계 명시""",
+                
+                AgentRole.CREATIVE: """[창의성생성자]
+입력구조 따라 창의적 확장
+��용예시+혁신접근+구체조언
+불필요설명 제거""",
+                
+                AgentRole.CRITIC: """[비평자-검증]
+신속검토: 정확성/논리성/실용성
+개선포인트 3개만
+각 2문장 이내""",
+                
+                AgentRole.FINALIZER: """[최종통합]
+모든의견 종합→최적답변
+명확구조+실용정보+창의균형
+바로 핵심 내용부터 시작. 불필요한 헤더나 마크업 없이. 마크다운 헤더(#, ##, ###) 사용 금지."""
+            },
+            'en': {
+                AgentRole.SUPERVISOR: """[Supervisor-Structure]
+Immediate analysis: core intent+required info+answer structure
+Output: 5 key points (1 sentence each)
+Clear reasoning framework""",
+                
+                AgentRole.CREATIVE: """[Creative Generator]
+Follow structure, expand creatively
+Practical examples+innovative approach+specific advice
+Remove unnecessary explanations""",
+                
+                AgentRole.CRITIC: """[Critic-Verification]
+Quick review: accuracy/logic/practicality
+Only 3 improvement points
+Max 2 sentences each""",
+                
+                AgentRole.FINALIZER: """[Final Integration]
+Synthesize all inputs→optimal answer
+Clear structure+practical info+creative balance
+Start with core content directly. No unnecessary headers or markup. No markdown headers (#, ##, ###)."""
+            },
+            'ja': {
+                AgentRole.SUPERVISOR: """[監督者-構造設計]
+即時分析：核心意図+必要情報+回答構造
+出力：5つの核心ポイント（各1文）
+推論体系明示""",
+                
+                AgentRole.CREATIVE: """[創造性生成者]
+入力構造に従って創造的拡張
+実用例+革新的アプローチ+具体的アドバイス
+不要な説明削除""",
+                
+                AgentRole.CRITIC: """[批評者-検証]
+迅速レビュー：正確性/論理性/実用性
+改善ポイント3つのみ
+各2文以内""",
+                
+                AgentRole.FINALIZER: """[最終統合]
+全意見統合→最適回答
+明確構造+実用情報+創造性バランス
+核心内容から直接開始。不要なヘッダーやマークアップなし。マークダウンヘッダー（#、##、###）使用禁止。"""
+            },
+            'zh': {
+                AgentRole.SUPERVISOR: """[主管-结构设计]
+立即分析：核心意图+所需信息+答案结构
+输出：5个核心要点（每个1句）
+推理体系明确""",
+                
+                AgentRole.CREATIVE: """[创意生成器]
+按结构创造性扩展
+实用示例+创新方法+具体建议
+删除不必要的解释""",
+                
+                AgentRole.CRITIC: """[评论家-验证]
+快速审查：准确性/逻辑性/实用性
+仅3个改进点
+每个最多2句""",
+                
+                AgentRole.FINALIZER: """[最终整合]
+综合所有意见→最佳答案
+清晰结构+实用信息+创意平衡
+直接从核心内容开始。无需不必要的标题或标记。禁止使用Markdown标题（#、##、###）。"""
+            }
+        }
+        
+        return prompts.get(lang, prompts['en'])
+    
+    async def parallel_process_agents(
+        self,
+        query: str,
+        search_results: List[Dict],
+        show_progress: bool = True,
+        lang: str = None
+    ) -> AsyncGenerator[Tuple[str, str], None]:
+        """병렬 처리 파이프라인 (캐싱 없음)"""
+        
+        start_time = time.time()
+        
+        # 언어 자동 감지
+        if lang is None:
+            lang = self.language_detector.detect_language(query)
+        
+        # 언어별 프롬프트 설정
+        self.compact_prompts = self._init_compact_prompts(lang)
+        
+        search_context = self._format_search_results(search_results)
+        accumulated_response = ""
+        agent_thoughts = ""
+        
+        # 추론 패턴 결정
+        reasoning_pattern = self.reasoning.get_reasoning_pattern(query, lang)
+        
+        try:
+            # === 1단계: 감독자 + 검색 병렬 실행 ===
+            if show_progress:
+                progress_msg = {
+                    'ko': "🚀 병렬 처리 시작\n👔 감독자 분석 + 🔍 추가 검색 동시 진행...\n\n",
+                    'en': "🚀 Starting parallel processing\n👔 Supervisor analysis + 🔍 Additional search in progress...\n\n",
+                    'ja': "🚀 並列処理開始\n👔 監督者分析 + 🔍 追加検索同時進行中...\n\n",
+                    'zh': "🚀 开始并行处理\n👔 主管分析 + 🔍 附加搜索同时进行...\n\n"
+                }
+                agent_thoughts = progress_msg.get(lang, progress_msg['en'])
+                yield accumulated_response, agent_thoughts
+            
+            # 감독자 프롬프트 (언어별)
+            supervisor_prompt_templates = {
+                'ko': f"""
+질문: {query}
+검색결과: {search_context}
+추론패턴: {reasoning_pattern}
+즉시 핵심구��� 5개 제시""",
+                'en': f"""
+Question: {query}
+Search results: {search_context}
+Reasoning pattern: {reasoning_pattern}
+Immediately provide 5 key structures""",
+                'ja': f"""
+質問: {query}
+検索結果: {search_context}
+推論パターン: {reasoning_pattern}
+即座に5つの核心構造を提示""",
+                'zh': f"""
+问题: {query}
+搜索结果: {search_context}
+推理模式: {reasoning_pattern}
+立即提供5个核心结构"""
+            }
+            
+            supervisor_prompt = supervisor_prompt_templates.get(lang, supervisor_prompt_templates['en'])
+            
+            supervisor_response = ""
+            supervisor_task = self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.SUPERVISOR]},
+                    {"role": "user", "content": supervisor_prompt}
+                ],
+                temperature=0.3,
+                max_tokens=500
+            )
+            
+            # 감독자 스트리밍 (버퍼링)
+            async for chunk in self.streaming.buffer_and_yield(supervisor_task):
+                supervisor_response += chunk
+                if show_progress and len(supervisor_response) < 300:
+                    supervisor_label = {
+                        'ko': "👔 감독자 분석",
+                        'en': "👔 Supervisor Analysis",
+                        'ja': "👔 監督者分析",
+                        'zh': "👔 主管分析"
+                    }
+                    agent_thoughts = f"{supervisor_label.get(lang, supervisor_label['en'])}\n{supervisor_response[:300]}...\n\n"
+                    yield accumulated_response, agent_thoughts
+            
+            # === 2단계: 창의성 + 비평 준비 병렬 ===
+            if show_progress:
+                creative_msg = {
+                    'ko': "🎨 창의성 생성자 + 🔍 비평자 준비...\n\n",
+                    'en': "🎨 Creative Generator + 🔍 Critic preparing...\n\n",
+                    'ja': "🎨 創造性生成者 + 🔍 批評者準備中...\n\n",
+                    'zh': "🎨 创意生成器 + 🔍 评论家准备中...\n\n"
+                }
+                agent_thoughts += creative_msg.get(lang, creative_msg['en'])
+                yield accumulated_response, agent_thoughts
+            
+            # 창의성 생성 시작 (언어별)
+            creative_prompt_templates = {
+                'ko': f"""
+질문: {query}
+감독자구조: {supervisor_response}
+검색결과: {search_context}
+창의적+실용적 답변 즉시생성""",
+                'en': f"""
+Question: {query}
+Supervisor structure: {supervisor_response}
+Search results: {search_context}
+Generate creative+practical answer immediately""",
+                'ja': f"""
+質問: {query}
+監督者構造: {supervisor_response}
+検索結果: {search_context}
+創造的+実用的回答即座生成""",
+                'zh': f"""
+问题: {query}
+主管结构: {supervisor_response}
+搜索结果: {search_context}
+立即生成创意+实用答案"""
+            }
+            
+            creative_prompt = creative_prompt_templates.get(lang, creative_prompt_templates['en'])
+            
+            creative_response = ""
+            creative_partial = ""
+            critic_started = False
+            critic_response = ""
+            
+            creative_task = self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.CREATIVE]},
+                    {"role": "user", "content": creative_prompt}
+                ],
+                temperature=0.8,
+                max_tokens=1500
+            )
+            
+            # 창의성 스트리밍 + 비평자 조기 시작
+            async for chunk in self.streaming.buffer_and_yield(creative_task):
+                creative_response += chunk
+                creative_partial += chunk
+                
+                # 창의성 응답이 500자 넘으면 비평자 시작
+                if len(creative_partial) > 500 and not critic_started:
+                    critic_started = True
+                    
+                    # 비평자 비동기 시작 (언어별)
+                    critic_prompt_templates = {
+                        'ko': f"""
+원본질문: {query}
+창의성답변(일부): {creative_partial}
+신속검토→개선점3개""",
+                        'en': f"""
+Original question: {query}
+Creative answer (partial): {creative_partial}
+Quick review→3 improvements""",
+                        'ja': f"""
+元の質問: {query}
+創造的回答（一部）: {creative_partial}
+迅速レビュー→改善点3つ""",
+                        'zh': f"""
+原始问题: {query}
+创意答案（部分）: {creative_partial}
+快速审查→3个改进点"""
+                    }
+                    
+                    critic_prompt = critic_prompt_templates.get(lang, critic_prompt_templates['en'])
+                    
+                    critic_task = asyncio.create_task(
+                        self._run_critic_async(critic_prompt)
+                    )
+                
+                if show_progress:
+                    display_creative = creative_response[:400] + "..." if len(creative_response) > 400 else creative_response
+                    creative_label = {
+                        'ko': "🎨 창의성 생성자",
+                        'en': "🎨 Creative Generator",
+                        'ja': "🎨 創造性生成者",
+                        'zh': "🎨 创意生成器"
+                    }
+                    agent_thoughts = f"{creative_label.get(lang, creative_label['en'])}\n{display_creative}\n\n"
+                    yield accumulated_response, agent_thoughts
+            
+            # 비평자 결과 대기
+            if critic_started:
+                critic_response = await critic_task
+                
+                if show_progress:
+                    critic_label = {
+                        'ko': "🔍 비평자 검토",
+                        'en': "🔍 Critic Review",
+                        'ja': "🔍 批評者レビュー",
+                        'zh': "🔍 评论家审查"
+                    }
+                    agent_thoughts += f"{critic_label.get(lang, critic_label['en'])}\n{critic_response[:200]}...\n\n"
+                    yield accumulated_response, agent_thoughts
+            
+            # === 3단계: 품질 체크 및 조기 종료 ===
+            quality_score, need_more = self.quality_checker.evaluate_response(
+                creative_response, query, lang
+            )
+            
+            if not need_more and quality_score > 0.85:
+                # 품질이 충분히 높으면 바로 반환
+                accumulated_response = self.response_cleaner.clean_response(creative_response)
+                
+                if show_progress:
+                    quality_msg = {
+                        'ko': f"✅ 품질 충족 (점수: {quality_score:.2f})\n조기 완료!\n",
+                        'en': f"✅ Quality met (score: {quality_score:.2f})\nEarly completion!\n",
+                        'ja': f"✅ 品質満足 (スコア: {quality_score:.2f})\n早期完了!\n",
+                        'zh': f"✅ 质量满足 (分数: {quality_score:.2f})\n提前完成!\n"
+                    }
+                    agent_thoughts += quality_msg.get(lang, quality_msg['en'])
+                
+                yield accumulated_response, agent_thoughts
+                return
+            
+            # === 4단계: 최종 통합 (스트리밍) ===
+            if show_progress:
+                final_msg = {
+                    'ko': "✅ 최종 통합 중...\n\n",
+                    'en': "✅ Final integration in progress...\n\n",
+                    'ja': "✅ 最終統合中...\n\n",
+                    'zh': "✅ 最终整合中...\n\n"
+                }
+                agent_thoughts += final_msg.get(lang, final_msg['en'])
+                yield accumulated_response, agent_thoughts
+            
+            # 최종 프롬프트 (언어별)
+            final_prompt_templates = {
+                'ko': f"""
+질문: {query}
+창의성답변: {creative_response}
+비평피드백: {critic_response}
+감독자구조: {supervisor_response}
+최종통합→완벽답변. 마크다운 헤더(#, ##, ###) 사용 금지.""",
+                'en': f"""
+Question: {query}
+Creative answer: {creative_response}
+Critic feedback: {critic_response}
+Supervisor structure: {supervisor_response}
+Final integration→perfect answer. No markdown headers (#, ##, ###).""",
+                'ja': f"""
+質問: {query}
+創造的回答: {creative_response}
+批評フィードバック: {critic_response}
+監督者構造: {supervisor_response}
+最終統合→完璧な回答。マークダウンヘッダー（#、##、###）使用禁止。""",
+                'zh': f"""
+问题: {query}
+创意答案: {creative_response}
+评论反馈: {critic_response}
+主管结构: {supervisor_response}
+最终整合→完美答案。禁止使用Markdown标题（#、##、###）。"""
+            }
+            
+            final_prompt = final_prompt_templates.get(lang, final_prompt_templates['en'])
+            
+            final_task = self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.FINALIZER]},
+                    {"role": "user", "content": final_prompt}
+                ],
+                temperature=0.5,
+                max_tokens=2500
+            )
+            
+            # 최종 답변 스트리밍
+            accumulated_response = ""
+            
+            async for chunk in final_task:
+                accumulated_response += chunk
+                # 실시간 정리
+                cleaned_response = self.response_cleaner.clean_response(accumulated_response)
+                yield cleaned_response, agent_thoughts
+            
+            # 최종 정리
+            accumulated_response = self.response_cleaner.clean_response(accumulated_response)
+            
+            # 처리 시간 추가 (언어별)
+            processing_time = time.time() - start_time
+            time_msg = {
+                'ko': f"\n\n---\n⚡ 처리 시간: {processing_time:.1f}초",
+                'en': f"\n\n---\n⚡ Processing time: {processing_time:.1f} seconds",
+                'ja': f"\n\n---\n⚡ 処理時間: {processing_time:.1f}秒",
+                'zh': f"\n\n---\n⚡ 处理时间: {processing_time:.1f}秒"
+            }
+            accumulated_response += time_msg.get(lang, time_msg['en'])
+            
+            yield accumulated_response, agent_thoughts
+            
+        except Exception as e:
+            error_msg = {
+                'ko': f"❌ 오류 발생: {str(e)}",
+                'en': f"❌ Error occurred: {str(e)}",
+                'ja': f"❌ エラー発生: {str(e)}",
+                'zh': f"❌ 发生错误: {str(e)}"
+            }
+            yield error_msg.get(lang, error_msg['en']), agent_thoughts
+    
+    async def _run_critic_async(self, prompt: str) -> str:
+        """비평자 비동기 실행 with error handling"""
+        try:
+            response = ""
+            async for chunk in self.llm.chat_stream_async(
+                messages=[
+                    {"role": "system", "content": self.compact_prompts[AgentRole.CRITIC]},
+                    {"role": "user", "content": prompt}
+                ],
+                temperature=0.2,
+                max_tokens=500
+            ):
+                response += chunk
+            return response
+        except Exception as e:
+            # 언어 감지
+            lang = 'ko' if '질문' in prompt else 'en'
+            error_msg = {
+                'ko': "비평 처리 중 오류",
+                'en': "Error during critic processing",
+                'ja': "批評処理中のエラー",
+                'zh': "评论处理中出错"
+            }
+            return error_msg.get(lang, error_msg['en'])
+    
+    def _format_search_results(self, results: List[Dict]) -> str:
+        """검색 결과 압축 포맷"""
+        if not results:
+            return "No search results"
+        
+        formatted = []
+        for i, r in enumerate(results[:3], 1):
+            title = r.get('title', '')[:50]
+            desc = r.get('description', '')[:100]
+            formatted.append(f"[{i}]{title}:{desc}")
+        
+        return " | ".join(formatted)
+
+
+# ============================================================================
+# Gradio UI (최적화 버전 - 캐싱 제거)
+# ============================================================================
+
+def create_optimized_gradio_interface():
+    """최적화된 Gradio 인터페이스 (캐싱 없음)"""
+    
+    # 시스템 초기화
+    system = SpeedOptimizedMultiAgentSystem()
+    
+    def process_query_optimized(
+        message: str,
+        history: List[Dict],
+        use_search: bool,
+        show_agent_thoughts: bool,
+        search_count: int,
+        language_mode: str
+    ):
+        """최적화된 쿼리 처리 - 실시간 스트리밍 버전"""
+        
+        if not message:
+            yield history, "", ""
+            return
+        
+        # 언어 설정
+        if language_mode == "Auto":
+            lang = None  # 자동 감지
+        else:
+            lang_map = {"Korean": "ko", "English": "en", "Japanese": "ja", "Chinese": "zh"}
+            lang = lang_map.get(language_mode, None)
+        
+        # 비동기 함수를 동기적으로 실행
+        try:
+            import nest_asyncio
+            nest_asyncio.apply()
+        except ImportError:
+            pass
+        
+        try:
+            # 검색 수행 (동기화)
+            search_results = []
+            search_display = ""
+            
+            # 언어 자동 감지 (필요한 경우)
+            detected_lang = lang or system.language_detector.detect_language(message)
+            
+            if use_search:
+                # 검색 상태 표시
+                processing_msg = {
+                    'ko': "⚡ 고속 처리 중...",
+                    'en': "⚡ High-speed processing...",
+                    'ja': "⚡ 高速処理中...",
+                    'zh': "⚡ 高速处理中..."
+                }
+                history_with_message = history + [
+                    {"role": "user", "content": message},
+                    {"role": "assistant", "content": processing_msg.get(detected_lang, processing_msg['en'])}
+                ]
+                yield history_with_message, "", ""
+                
+                # 비동기 검색을 동기적으로 실행
+                async def search_wrapper():
+                    return await system.search.search_async(message, count=search_count, lang=detected_lang)
+                
+                loop = asyncio.new_event_loop()
+                asyncio.set_event_loop(loop)
+                search_results = loop.run_until_complete(search_wrapper())
+                
+                if search_results:
+                    ref_label = {
+                        'ko': "📚 참고 자료",
+                        'en': "📚 References",
+                        'ja': "📚 参考資料",
+                        'zh': "📚 参考资料"
+                    }
+                    search_display = f"{ref_label.get(detected_lang, ref_label['en'])}\n\n"
+                    for i, result in enumerate(search_results[:3], 1):
+                        search_display += f"**{i}. [{result['title'][:50]}]({result['url']})**\n"
+                        search_display += f"   {result['description'][:100]}...\n\n"
+            
+            # 사용자 메시지 추가
+            current_history = history + [{"role": "user", "content": message}]
+            
+            # 실시간 스트리밍을 위한 비동기 처리
+            async def stream_responses():
+                """실시간 스트리밍 제너레이터"""
+                async for response, thoughts in system.parallel_process_agents(
+                    query=message,
+                    search_results=search_results,
+                    show_progress=show_agent_thoughts,
+                    lang=detected_lang
+                ):
+                    yield response, thoughts
+            
+            # 새 이벤트 루프에서 실시간 스트리밍
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+            
+            # 비동기 제너레이터를 동기적으로 순회
+            gen = stream_responses()
+            
+            while True:
+                try:
+                    # 다음 항목 가져오기
+                    task = asyncio.ensure_future(gen.__anext__(), loop=loop)
+                    response, thoughts = loop.run_until_complete(task)
+                    
+                    # 실시간 업데이트
+                    updated_history = current_history + [
+                        {"role": "assistant", "content": response}
+                    ]
+                    yield updated_history, thoughts, search_display
+                    
+                except StopAsyncIteration:
+                    break
+            
+        except Exception as e:
+            error_history = history + [
+                {"role": "user", "content": message},
+                {"role": "assistant", "content": f"❌ Error: {str(e)}"}
+            ]
+            yield error_history, "", ""
+        finally:
+            # 루프 정리
+            try:
+                loop.close()
+            except:
+                pass
+    
+    # Gradio 인터페이스
+    with gr.Blocks(
+        title="⚡ Speed-Optimized Multi-Agent System (No Cache)",
+        theme=gr.themes.Soft(),
+        css="""
+        .gradio-container {
+            max-width: 1400px !important;
+            margin: auto !important;
+        }
+        """
+    ) as demo:
+        gr.Markdown("""
+        # ⚡ Enhanced Multi-Agent RAG System (캐싱 제거 버전)
+        **Complex questions processed within 5-8 seconds | Multi-language support**
+        
+        **Optimization Features:**
+        - 🚀 Parallel Processing: Concurrent agent execution
+        - ⚡ Stream Buffering: Network optimization
+        - 🎯 Early Termination: Complete immediately when quality is met
+        - 🌍 Multi-language: Auto-detect Korean/English/Japanese/Chinese
+        - ❌ **Caching Disabled**: 캐싱 기능 제거됨
+        """)
+        
+        with gr.Row():
+            with gr.Column(scale=3):
+                chatbot = gr.Chatbot(
+                    height=500,
+                    label="💬 Chat",
+                    type="messages"
+                )
+                
+                msg = gr.Textbox(
+                    label="Enter complex question",
+                    placeholder="Enter complex questions requiring analysis, strategy, or creative solutions...",
+                    lines=3
+                )
+                
+                with gr.Row():
+                    submit = gr.Button("⚡ High-Speed Process", variant="primary")
+                    clear = gr.Button("🔄 Reset")
+                
+                with gr.Accordion("🤖 Agent Processing", open=False):
+                    agent_thoughts = gr.Markdown()
+                
+                with gr.Accordion("📚 Search Sources", open=False):
+                    search_sources = gr.Markdown()
+            
+            with gr.Column(scale=1):
+                gr.Markdown("**⚙️ Settings**")
+                
+                language_mode = gr.Radio(
+                    choices=["Auto", "Korean", "English", "Japanese", "Chinese"],
+                    value="Auto",
+                    label="🌍 Language Mode"
+                )
+                
+                use_search = gr.Checkbox(
+                    label="🔍 Use Web Search",
+                    value=True
+                )
+                
+                show_agent_thoughts = gr.Checkbox(
+                    label="🧠 Show Processing",
+                    value=True
+                )
+                
+                search_count = gr.Slider(
+                    minimum=3,
+                    maximum=10,
+                    value=5,
+                    step=1,
+                    label="Search Results Count"
+                )
+                
+                gr.Markdown("""
+                **⚡ Optimization Status**
+                
+                **Active Optimizations:**
+                - ✅ Parallel Processing
+                - ❌ ~~Smart Caching~~ (제거됨)
+                - ✅ Buffer Streaming
+                - ✅ Early Termination
+                - ✅ Compressed Prompts
+                - ✅ Multi-language Support
+                - ✅ Error Recovery
+                
+                **Expected Processing Time:**
+                - Simple Query: 3-5 seconds
+                - Complex Query: 5-8 seconds
+                - Very Complex: 8-12 seconds
+                """)
+        
+        # 복잡한 질문 예제 (다국어)
+        gr.Examples(
+            examples=[
+                # Korean
+                "AI 기술이 향후 10년간 한국 경제에 미칠 영향을 다각도로 분석하고 대응 전략을 제시해줘",
+                "스타트업이 대기업과 경쟁하기 위한 혁신적인 전략을 단계별로 수립해줘",
+                # English
+                "Analyze the multifaceted impact of quantum computing on current encryption systems and propose alternatives",
+                "Design 5 innovative business models for climate change mitigation with practical implementation details",
+                # Japanese
+                "メタバース時代の教育革新方案を実装可能なレベルで提案してください",
+                # Chinese
+                "分析人工智能对未来十年全球经济的影响并提出应对策略"
+            ],
+            inputs=msg
+        )
+        
+        # 이벤트 바인딩
+        submit.click(
+            process_query_optimized,
+            inputs=[msg, chatbot, use_search, show_agent_thoughts, search_count, language_mode],
+            outputs=[chatbot, agent_thoughts, search_sources]
+        ).then(
+            lambda: "",
+            None,
+            msg
+        )
+        
+        msg.submit(
+            process_query_optimized,
+            inputs=[msg, chatbot, use_search, show_agent_thoughts, search_count, language_mode],
+            outputs=[chatbot, agent_thoughts, search_sources]
+        ).then(
+            lambda: "",
+            None,
+            msg
+        )
+        
+        clear.click(
+            lambda: ([], "", ""),
+            None,
+            [chatbot, agent_thoughts, search_sources]
+        )
+    
+    return demo
+
+
+# ============================================================================
+# 메인 실행
+# ============================================================================
+
+if __name__ == "__main__":
+    print("""
+╔══════════════════════════════════════════════════════════════╗
+║     ⚡ Speed-Optimized Multi-Agent System (No Cache) ⚡     ║
+║                                                              ║
+║   High-speed AI system processing complex questions         ║
+║                                                              ║
+║  Features:                                                  ║
+║  • Multi-language support (KO/EN/JA/ZH)                    ║
+║  • Improved error recovery                                  ║
+║  • NO CACHING (캐싱 기능 제거됨)                           ║
+║  • Adaptive stream buffering                                ║
+║  • Response cleaning & formatting                           ║
+╚══════════════════════════════════════════════════════════════╝
+    """)
+    
+    # API 키 확인
+    if not os.getenv("FIREWORKS_API_KEY"):
+        print("\n⚠️  FIREWORKS_API_KEY is not set.")
+    
+    if not os.getenv("BRAVE_SEARCH_API_KEY"):
+        print("\n⚠️  BRAVE_SEARCH_API_KEY is not set.")
+    
+    # Gradio 앱 실행
+    demo = create_optimized_gradio_interface()
+    
+    is_hf_spaces = os.getenv("SPACE_ID") is not None
+    
+    if is_hf_spaces:
+        print("\n🤗 Running in optimized mode on Hugging Face Spaces (No Cache)...")
+        demo.launch(server_name="0.0.0.0", server_port=7860)
+    else:
+        print("\n💻 Running in optimized mode on local environment (No Cache)...")
+        demo.launch(server_name="0.0.0.0", server_port=7860, share=False), '', response, flags=re.MULTILINE)
+        
+        # 리스트 마커를 일반 텍스트로 변경
+        response = re.sub(r'^\s*[\*\-\+]\s+', '• ', response, flags=re.MULTILINE)
+        response = re.sub(r'^\s*\d+\.\s+', '', response, flags=re.MULTILINE)
+        
+        # 링크 텍스트만 남기기 [text](url) -> text
+        response = re.sub(r'\[([^\]]+)\]\([^\)]+\)', r'\1', response)
+        
+        # 이미지 제거 ![alt](url)
+        response = re.sub(r'!\[([^\]]*)\]\([^\)]+\)', '', response)
         
         # 중복 공백 제거
         response = re.sub(r'\n{3,}', '\n\n', response)
+        response = re.sub(r' {2,}', ' ', response)
         
         # 특정 패턴 제거
         unwanted_patterns = [
@@ -486,12 +2792,18 @@ class ResponseCleaner:
             r'^\s*\*\*\[.*?\]\*\*\s*',  # [태그] 형식 제거
             r'^\s*###\s*',  # ### 제거
             r'^\s*##\s*',   # ## 제거
-            r'^\s*#\s*'     # # 제거
+            r'^\s*#\s*',    # # 제거
+            r'^\s*>\s+',    # 인용문 마커 제거
         ]
         
         for pattern in unwanted_patterns:
             response = re.sub(pattern, '', response, flags=re.MULTILINE)
         
+        # 빈 줄 정리
+        lines = [line.strip() for line in response.split('\n')]
+        lines = [line for line in lines if line]
+        response = '\n\n'.join(lines)
+        
         return response.strip()