app.py

import os
import gradio as gr
import logging
from typing import List, Tuple
from graphrag_agent import (
    build_graphrag_agent,
    system_initialized,
    MyState
)
from langchain_core.messages import HumanMessage

# ======================== 配置 ========================
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# ======================== 全局变量 ========================
compiled_graph = None

# 会话状态存储：用于跟踪每个用户的对话状态
session_states = {}

# ======================== 初始化系统 ========================
def check_system_status():
    """检查系统初始化状态"""
    global compiled_graph
    
    try:
        # 导入 graphrag_agent 已经自动加载了数据
        # 我们只需要构建图
        if system_initialized and compiled_graph is None:
            logger.info("🔧 Building GraphRAG Agent...")
            compiled_graph = build_graphrag_agent()
            logger.info("✅ GraphRAG Agent compiled successfully!")
            return "✅ System ready! You can start asking questions."
        elif system_initialized:
            return "✅ System already initialized."
        else:
            return "⏳ System is still initializing, please wait..."
    except Exception as e:
        logger.error(f"❌ Initialization check failed: {e}")
        return f"❌ Error: {str(e)}"

# ======================== 核心处理函数 ========================
def process_message(message: str, history: List[Tuple[str, str]], session_id: str):
    """
    处理用户消息的核心函数
    
    Args:
        message: 用户输入的消息
        history: 对话历史 [(user_msg, bot_msg), ...]
        session_id: 会话ID，用于跟踪状态
    
    Returns:
        ("", updated_history, session_state_display)
    """
    
    # 检查系统是否初始化
    if not system_initialized or compiled_graph is None:
        history.append((message, "⚠️ System not initialized. Please wait for data loading..."))
        return "", history, "System Status: Not Ready"
    
    if not message.strip():
        return "", history, "System Status: Waiting for input"
    
    try:
        # 获取或创建会话状态
        if session_id not in session_states:
            session_states[session_id] = {
                "waiting_for_clarification": False,
                "current_graph_state": None,
                "conversation_messages": []
            }
        
        session_state = session_states[session_id]
        
        # ===== 情况 1: 正在等待用户补充信息 =====
        if session_state["waiting_for_clarification"]:
            logger.info(f"📝 User provided clarification: {message}")
            
            # 将用户的补充信息添加到对话历史
            session_state["conversation_messages"].append(HumanMessage(content=message))
            
            # 从 user_input 节点继续执行
            prev_state = session_state["current_graph_state"]
            
            # 调用 user_input 节点，传入用户回复
            from graphrag_agent import user_input
            user_input_result = user_input(prev_state, user_reply_text=message)
            
            # 更新状态
            prev_state.update(user_input_result)
            
            # 如果还需要更多信息（理论上不会，但以防万一）
            if user_input_result.get("need_user_reply", False):
                ai_msg = user_input_result.get("ai_message", "Please provide more information.")
                history.append((message, f"🤔 {ai_msg}"))
                return "", history, "Status: Still waiting for clarification"
            
            # 继续执行图的剩余部分（从 parse_query 重新开始）
            result = compiled_graph.invoke(prev_state)
            
            # 获取最终答案
            final_answer = result.get("llm_answer", "No answer generated.")
            history.append((message, final_answer))
            
            # 重置等待状态
            session_state["waiting_for_clarification"] = False
            session_state["current_graph_state"] = None
            
            return "", history, "✅ Status: Answer generated (after clarification)"
        
        # ===== 情况 2: 新的查询 =====
        else:
            logger.info(f"🔍 Processing new query: {message}")
            
            # 将用户消息添加到会话历史
            session_state["conversation_messages"].append(HumanMessage(content=message))
            
            # 创建初始状态
            initial_state = {
                "messages": session_state["conversation_messages"].copy()
            }
            
            # 执行图 - 使用流式处理以捕获中间状态
            final_state = None
            need_clarification = False
            clarification_message = None
            
            for event in compiled_graph.stream(initial_state):
                # 检查是否到达 user_input 节点
                if "user_input" in event:
                    user_input_state = event["user_input"]
                    
                    # 检查是否需要用户补充信息
                    if user_input_state.get("need_user_reply", False):
                        need_clarification = True
                        clarification_message = user_input_state.get("ai_message", "Please provide more information.")
                        
                        # 保存当前状态，等待用户回复
                        session_state["waiting_for_clarification"] = True
                        session_state["current_graph_state"] = user_input_state
                        
                        # 不继续执行，等待用户输入
                        break
                
                # 保存最终状态
                final_state = event
            
            # 如果需要用户补充信息
            if need_clarification:
                history.append((message, f"🤔 **Need More Information:**\n\n{clarification_message}\n\n*Please provide additional details.*"))
                return "", history, "⏳ Status: Waiting for clarification"
            
            # 如果直接得到答案（信息充足的情况）
            if final_state:
                # 获取最终答案
                answer_key = list(final_state.keys())[0] if final_state else None
                if answer_key:
                    state_data = final_state[answer_key]
                    final_answer = state_data.get("llm_answer", "No answer generated.")
                else:
                    final_answer = "Unable to generate answer."
                
                history.append((message, final_answer))
                return "", history, "✅ Status: Answer generated"
            else:
                history.append((message, "❌ Error: No response from agent."))
                return "", history, "❌ Status: Error"
    
    except Exception as e:
        logger.error(f"❌ Error processing message: {e}")
        import traceback
        traceback.print_exc()
        
        error_msg = f"❌ Error: {str(e)}\n\nPlease try again or rephrase your question."
        history.append((message, error_msg))
        
        # 重置会话状态
        if session_id in session_states:
            session_states[session_id]["waiting_for_clarification"] = False
        
        return "", history, f"❌ Status: Error - {str(e)}"

# ======================== 清除会话 ========================
def clear_session(session_id: str):
    """清除会话状态"""
    if session_id in session_states:
        del session_states[session_id]
    return [], "🔄 Status: Session cleared"

# ======================== Gradio 界面 ========================
def create_demo():
    """创建 Gradio 界面"""
    
    with gr.Blocks(
        theme=gr.themes.Soft(),
        title="Dental Knowledge QA System",
        css="""
        .chatbot-container {
            height: 600px !important;
        }
        .status-box {
            background-color: #f0f0f0;
            padding: 10px;
            border-radius: 5px;
            font-family: monospace;
        }
        """
    ) as demo:
        
        # 标题
        gr.Markdown("""
        # 🦷 Dental Knowledge QA System
        ### GraphRAG-based Agent with Knowledge Graph & External Search
        
        Ask any dental medicine question. The system will:
        1. Parse your query and extract entities
        2. **Ask for clarification if needed** 🔄
        3. Search the knowledge graph
        4. Optionally search PubMed & Wikipedia
        5. Generate a comprehensive answer with sources
        """)
        
        # 系统状态
        with gr.Row():
            init_status = gr.Textbox(
                label="🔧 System Status",
                value="Checking system status...",
                interactive=False,
                lines=2
            )
        
        # 主对话区域
        with gr.Row():
            with gr.Column(scale=4):
                chatbot = gr.Chatbot(
                    label="💬 Conversation",
                    height=600,
                    bubble_full_width=False,
                    avatar_images=(None, "🤖")
                )
                
                with gr.Row():
                    msg_input = gr.Textbox(
                        label="Your Question",
                        placeholder="Ask a dental question... (e.g., 'What is the treatment of gingivitis?')",
                        lines=2,
                        scale=4
                    )
                    with gr.Column(scale=1):
                        send_btn = gr.Button("Send 📤", variant="primary", size="lg")
                        clear_btn = gr.Button("Clear 🗑️", size="lg")
                
                # 示例问题
                gr.Examples(
                    examples=[
                        "What is the treatment of gingivitis?",
                        "What are the complications of implant restoration?",
                        "How to prevent dental caries in children?",
                        "What is the best treatment?",  # 这个会触发交互
                        "印模材料凝固后，其软度通常用什么指标表示？"
                    ],
                    inputs=msg_input,
                    label="📝 Example Questions"
                )
            
            # 右侧状态面板
            with gr.Column(scale=1):
                session_status = gr.Textbox(
                    label="📊 Session Status",
                    value="Status: Ready",
                    interactive=False,
                    lines=3
                )
                
                gr.Markdown("""
                ### 💡 Tips
                - Ask clear, specific questions
                - If asked for clarification, provide more details
                - Include context when possible
                
                ### 🔄 Interaction Flow
                1. **Insufficient info** → System asks for clarification
                2. You provide details → System continues
                3. **Sufficient info** → Direct answer
                """)
        
        # 技术信息
        with gr.Accordion("ℹ️ System Information", open=False):
            gr.Markdown("""
            ### Architecture
            - **Knowledge Graph**: Neo4j-based dental knowledge
            - **Retrieval**: FAISS + SapBERT embeddings
            - **Reranking**: BGE bi-encoder + cross-encoder
            - **LLM**: DeepSeek Chat
            - **External Sources**: PubMed, Wikipedia
            
            ### Features
            - Multi-turn conversation support
            - Intelligent query clarification
            - Entity extraction and linking
            - Source attribution
            """)
        
        # 隐藏的会话ID（每个用户独立）
        session_id = gr.State(value=lambda: str(id(object())))
        
        # ===== 事件绑定 =====
        
        # 检查系统状态（页面加载时）
        demo.load(
            fn=check_system_status,
            outputs=init_status
        )
        
        # 发送消息
        send_btn.click(
            fn=process_message,
            inputs=[msg_input, chatbot, session_id],
            outputs=[msg_input, chatbot, session_status]
        )
        
        # 回车发送
        msg_input.submit(
            fn=process_message,
            inputs=[msg_input, chatbot, session_id],
            outputs=[msg_input, chatbot, session_status]
        )
        
        # 清除对话
        clear_btn.click(
            fn=clear_session,
            inputs=session_id,
            outputs=[chatbot, session_status]
        )
    
    return demo

# ======================== 启动应用 ========================
if __name__ == "__main__":
    demo = create_demo()
    demo.queue(max_size=20)  # 支持并发
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False,
        show_error=True
    )