Update app.py

app.py

@@ -1,16 +1,22 @@
 #!/usr/bin/env python
 import os
 import shutil
+import tempfile
 import json
 import torch
 import re
 import requests
 import transformers
 import chardet
+import deepeval
+import difflib
 from transformers import AutoModelForCausalLM, AutoTokenizer
 from transformers.models.llama.configuration_llama import LlamaConfig
 from huggingface_hub import hf_hub_download
+from typing import List, Dict, Any
 import gradio as gr
+from pathlib import Path
+
 
 # Solve permission issues
 os.environ["MPLCONFIGDIR"] = "/tmp/matplotlib"
@@ -24,6 +30,44 @@ os.environ["GRADIO_FLAGGING_DIR"] = "/tmp/flagged"
 os.environ["SENTENCE_TRANSFORMERS_HOME"] = "/tmp/sentence_transformers"
 os.environ["HF_HUB_CACHE"] = "/tmp/huggingface/hf_cache"
 os.environ["HF_HUB_DOWNLOAD_TIMEOUT"] = "60"
+os.environ["LANGCHAIN_TRACING_V2"] = "true"
+os.environ["LANGCHAIN_API_KEY"] = os.getenv("LANGCHAIN_API_KEY")
+os.environ["LANGCHAIN_PROJECT"] = os.getenv("LANGCHAIN_PROJECT")
+
+# 設置環境變數，確保 AutoGen 可以寫入臨時目錄
+os.environ["AUTOGEN_WORKSPACE"] = "/tmp/autogen_workspace"
+os.makedirs("/tmp/autogen_workspace", exist_ok=True)
+os.chmod("/tmp/autogen_workspace", 0o777)  # 確保目錄可寫
+
+# 設置 OpenAI API 相關環境變數
+os.environ["OPENAI_API_TYPE"] = "open_ai"  # 如果您使用的是 OpenAI API
+
+
+
+# ✅ 建立 temp 安全區
+os.environ["DEEPEVAL_TELEMETRY_OPT_OUT"] = "YES"
+os.environ["DEEPEVAL_RESULTS_FOLDER"] = "/tmp/deepeval_results"
+os.makedirs("/tmp/deepeval_results", exist_ok=True)
+
+# ✅ 修正 Python tempdir 基底（避免它寫 home）
+import tempfile
+tempfile.tempdir = "/tmp"
+# 在此處加入 DeepEval 的 monkey-patch，避免全域更改工作目錄
+original_evaluate = deepeval.evaluate
+
+def patched_evaluate(*args, **kwargs):
+    current_dir = os.getcwd()
+    try:
+        os.chdir("/tmp")
+        return original_evaluate(*args, **kwargs)
+    finally:
+        os.chdir(current_dir)
+
+deepeval.evaluate = patched_evaluate
+
+
+SHOW_EVAL = os.getenv("SHOW_EVAL", "false").lower() == "true"
+
 
 # Load Required Modules 
 from langchain.embeddings import HuggingFaceEmbeddings
@@ -53,7 +97,26 @@ from crewai.tools import tool
 from geopy.geocoders import Nominatim
 from timezonefinder import TimezoneFinder
 from langchain_experimental.agents import create_pandas_dataframe_agent
-
+from langsmith import traceable
+from deepeval import evaluate
+from deepeval.metrics import AnswerRelevancyMetric
+from deepeval.test_case import LLMTestCase
+# from langgraph.graph import Graph
+from langgraph.graph import StateGraph
+from langchain_core.runnables import RunnableLambda
+from langchain.chains import LLMChain
+from langchain.chains.combine_documents.stuff import StuffDocumentsChain
+from sentence_transformers import SentenceTransformer
+# === AutoGen for multi-intent collaboration ===
+from autogen import AssistantAgent, UserProxyAgent, GroupChat, GroupChatManager
+
+
+
+try:
+    from phoenix.trace.langchain import LangChainInstrumentor
+    LangChainInstrumentor().instrument()
+except Exception as e:
+    print(f"[WARNING] Failed to load Phoenix LangChain trace: {e}")
 
 session_retriever = None
 session_qa_chain = None
@@ -193,7 +256,8 @@ def get_file_path(file):
     else:
         return None
 
-# Original functionalities (Tabs 1-4) functions 
+# Original functionalities (Tabs 1-4) functions
+@traceable(name="Biden LLaMA QA")
 def rag_llama_qa(query):
     output = RetrievalQA.from_chain_type(
         llm=llm_local,
@@ -206,9 +270,30 @@ def rag_llama_qa(query):
     idx = lower_text.find("answer:")
     return output[idx + len("answer:"):].strip() if idx != -1 else output
 
+@traceable(name="GPT-4 Document QA")
 def rag_gpt4_qa(query):
-    return qa_gpt.run(query)
+    raw_answer = qa_gpt.run(query)
+
+    if SHOW_EVAL:
+        try:
+            top_docs = retriever.get_relevant_documents(query)
+            test_case = LLMTestCase(
+                input=query,
+                actual_output=raw_answer,
+                expected_output=raw_answer,
+                context=[doc.page_content for doc in top_docs[:3]]
+            )
+            metric = AnswerRelevancyMetric(model="gpt-4o-mini")
+            results = evaluate([test_case], [metric])
+            result = results[0]
+            print(f"[DeepEval Tab4] Input: {query}")
+            print(f"[DeepEval Tab4] Passed: {result.passed}, Score: {result.score:.2f}, Reason: {result.reason}")
+        except Exception as e:
+            print(f"[DeepEval Tab4] Evaluation failed: {e}")
+
+    return raw_answer
 
+@traceable(name="Upload Document QA")
 def upload_and_chat(file, query):
     file_path = get_file_path(file)
     if file_path is None:
@@ -230,8 +315,26 @@ def upload_and_chat(file, query):
         return_source_documents=False,
         chain_type_kwargs={"prompt": custom_prompt}
     )
-    return qa_temp.run(query)
+    raw_answer = qa_temp.run(query)
+    if SHOW_EVAL:
+        try:
+            test_case = LLMTestCase(
+                input=query,
+                actual_output=raw_answer,
+                expected_output=raw_answer,
+                context=[d.page_content for d in chunks[:3]]
+            )
+            metric = AnswerRelevancyMetric(model="gpt-4o-mini")  # default is GPT-4o
+            results = evaluate([test_case], [metric])
+            result = results[0]
+            print(f"[DeepEval QA] Input: {query}")
+            print(f"[DeepEval QA] Passed: {result.passed}, Score: {result.score:.2f}, Reason: {result.reason}")
+        except Exception as e:
+            print(f"[DeepEval QA] Evaluation failed: {e}")
 
+    return raw_answer
+
+   
 initial_prompt = PromptTemplate(
     input_variables=["text"],
     template="""Write a concise and structured summary of the following content. Focus on capturing the main ideas and key details:
@@ -254,6 +357,7 @@ Refine the summary based on the new content below. Add or update information onl
 """
 )
 
+@traceable(name="Document Summarise")
 def document_summarize(file):
     file_path = get_file_path(file)
     if file_path is None:
@@ -453,7 +557,7 @@ def time_tool(query: str) -> str:
         location = "London"
 
     location_key = location.lower()
-    tz_str = zone_map.get(location_key, "Europe/London")
+    tz_str = location_to_timezone(location)
     now = datetime.now(ZoneInfo(tz_str))
 
     # return time or date
@@ -982,6 +1086,49 @@ crew = Crew(
     llm=crew_llm
 )
 
+# test qa
+def build_langgraph_doc_qa_chain(llm, retriever, memory, prompt):
+    def retrieve_step(state):
+        docs = state['retriever'].get_relevant_documents(state['query'])
+        return {"docs": docs, **state}
+
+    def answer_step(state):
+        prompt = state["prompt"]
+        llm = state["llm"]
+        docs = state["docs"]
+
+        llm_chain = LLMChain(llm=llm, prompt=prompt)
+        doc_chain = StuffDocumentsChain(
+            llm_chain=llm_chain,
+            document_variable_name="context"
+        )
+    # 只執行一次，並傳入所有需要的參數
+        answer = doc_chain.run({
+            "input_documents": docs,
+            "question": state["query"]
+        })
+        return {"answer": answer, **state}
+
+    builder = StateGraph(dict)
+    builder.add_node("Retrieve", retrieve_step)
+    builder.add_node("Answer", answer_step)
+    builder.set_entry_point("Retrieve")
+    builder.add_edge("Retrieve", "Answer")
+    builder.set_finish_point("Answer")
+
+    compiled = builder.compile()
+
+    def run(query):
+        return compiled.invoke({
+            "query": query,
+            "retriever": retriever,
+            "llm": llm,
+            "prompt": prompt
+        })
+
+    return run
+
+@traceable(name="Multi-Agent Chat")
 def multi_agent_chat_advanced(query: str, file=None) -> str:
     global session_retriever, session_qa_chain, csv_dataframe
 
@@ -1060,7 +1207,28 @@ def multi_agent_chat_advanced(query: str, file=None) -> str:
                 # If using QA Chain is appropriate
                 if use_file_chain:
                     try:
-                        return session_qa_chain.run(query)
+                        answer = session_qa_chain.run(query)
+                        #session_graph_chain = build_langgraph_doc_qa_chain(llm_gpt4, session_retriever, memory, custom_prompt)
+                        #answer = session_graph_chain(query)["answer"]
+
+                        # ✅ DeepEval 評估僅在 Tab1 文件 QA 的情況下觸發
+                        if SHOW_EVAL:
+                            try:
+                                test_case = LLMTestCase(
+                                    input=query,
+                                    actual_output=answer,
+                                    expected_output=answer,
+                                    context=[d.page_content for d in session_retriever.get_relevant_documents(query)[:3]]
+                                )
+                                metric = AnswerRelevancyMetric(model="gpt-4o-mini")
+                                results = evaluate([test_case], [metric])
+                                result = results[0]
+                                print(f"[DeepEval Tab1] Input: {query}")
+                                print(f"[DeepEval Tab1] Passed: {result.passed}, Score: {result.score:.2f}, Reason: {result.reason}")
+                            except Exception as e:
+                                print(f"[DeepEval Tab1] Evaluation failed: {e}")
+
+                        return answer
                     except Exception as e:
                         return f"Document QA Error: {e}"
 
@@ -1084,6 +1252,471 @@ def multi_agent_chat_advanced(query: str, file=None) -> str:
     except Exception as e:
         return f"Multi-Agent Error: {e}"
 
+
+
+# LangGraph 使用的節點函數（會接續你的 Crew Agent）
+# 初始化 embedding model
+embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
+
+# Intent Embedding 分類（支援檔名）
+INTENT_LABELS = {
+    "DocQA": ["document", "file", "paper", "cb", "proposal", "project"],
+    "Summarise": ["summarise", "summary", "abstract", "key points", "overview", "main points"],
+    "General": ["who are you", "tell me something", "what can you do", "fun fact"],
+}
+
+def parse_query(query: str) -> dict:
+    prompt = """Analyze the following query and determine required subtasks. Return a JSON object containing:
+    - summarize_files: list of document indices to summarize
+    - qa_pairs: list of QA objects [{"question": "question", "doc_indices": [relevant doc indices]}]
+    - compare_files: list of document index pairs to compare [[doc1_idx, doc2_idx]]
+    - find_relations: boolean, whether to analyze document relationships
+    
+    For example, query "What are the differences between document A and B, and summarize A" should return:
+    {
+        "summarize_files": [0],
+        "qa_pairs": [],
+        "compare_files": [[0, 1]],
+        "find_relations": false
+    }
+    
+    Query: """ + query
+    
+    response = llm_gpt4.invoke(prompt)
+    try:
+        return json.loads(response.content)
+    except:
+        return {
+            "summarize_files": [],
+            "qa_pairs": [{"question": query, "doc_indices": [0]}],
+            "compare_files": [],
+            "find_relations": False
+        }
+
+
+def autogen_multi_document_analysis(query: str, docs: list, file_names: list) -> str:
+    try:
+        # 建立絕對路徑的暫存目錄，並確保它存在
+        import tempfile
+        import os
+        
+        # 建立一個臨時工作目錄
+        temp_dir = tempfile.mkdtemp(dir="/tmp")
+        os.environ["OPENAI_CACHE_DIR"] = temp_dir
+        
+        # 設置 AutoGen 的工作目錄
+        os.environ["AUTOGEN_CACHE_PATH"] = temp_dir
+        os.environ["AUTOGEN_CACHEDIR"] = temp_dir
+        os.environ["OPENAI_CACHE_PATH"] = temp_dir
+        
+        # 強制 AutoGen 使用我們的臨時目錄而不是 ./.cache
+        import autogen
+        if hasattr(autogen, "set_cache_dir"):
+            autogen.set_cache_dir(temp_dir)
+        
+        # 準備文件上下文
+        context = "\n\n".join(
+            f"Document {name}:\n{doc[:2000]}..." 
+            for name, doc in zip(file_names, docs)
+        )
+
+        # 配置 LLM
+        config_list = [{
+            "model": "gpt-4o-mini",
+            "api_key": openai_api_key
+        }]
+
+        # 基礎配置 - 不包含任何緩存相關參數
+        llm_config = {
+            "config_list": config_list,
+            "temperature": 0
+        }
+        
+        # 在進行 AutoGen 處理前，切換到臨時目錄
+        original_dir = os.getcwd()
+        os.chdir(temp_dir)
+        
+        try:
+            # 以下是您的 AutoGen 處理代碼
+            user_proxy = UserProxyAgent(
+                name="User",
+                system_message="A user seeking information from multiple documents.",
+                human_input_mode="NEVER",
+                code_execution_config={"use_docker": False},
+                llm_config=llm_config
+            )
+
+
+
+            # 定義文檔分析專家
+            doc_analyzer = AssistantAgent(
+                name="DocumentAnalyzer",
+                system_message="""You are an expert at analyzing and comparing documents. Focus on:
+                1. Key similarities and differences
+                2. Main themes and topics
+                3. Relationships between documents
+                4. Evidence-based analysis""",
+                llm_config=llm_config
+            )
+
+            # 定義問答專家
+            qa_expert = AssistantAgent(
+                name="QAExpert",
+                system_message="""You are an expert at extracting specific information. Focus on:
+                1. Finding relevant details
+                2. Answering specific questions
+                3. Cross-referencing information
+                4. Providing evidence""",
+                llm_config=llm_config
+            )
+
+            # 定義總結專家
+            summarizer = AssistantAgent(
+                name="Summarizer",
+                system_message="""You are an expert at summarizing content. Focus on:
+                1. Key points and findings
+                2. Important relationships
+                3. Critical conclusions
+                4. Comprehensive overview""",
+                llm_config=llm_config
+            )
+
+            # 創建群組聊天
+            groupchat = GroupChat(
+                agents=[user_proxy, doc_analyzer, qa_expert, summarizer],
+                messages=[],
+                max_round=5
+            )
+
+            # 創建管理器
+            manager = GroupChatManager(
+                groupchat=groupchat,
+                llm_config=llm_config
+            )
+
+            # 準備任務提示
+            task_prompt = f"""Analyze these documents and answer the query:
+        
+            Query: {query}
+
+            Documents Context:
+            {context}
+
+            Requirements:
+            1. Provide a direct and clear answer
+            2. Support all claims with evidence from the documents
+            3. Consider relationships between all documents
+            4. If comparing, analyze all relevant aspects
+            5. If summarizing, cover all important points
+            6. If looking for specific content, search thoroughly
+            7. If analyzing relationships, consider all connections
+
+            Please provide a comprehensive and well-structured answer."""
+
+            # 執行群組討論
+            user_proxy.initiate_chat(manager, message=task_prompt)
+            return user_proxy.last_message()["content"]
+        finally:
+            # 完成後，切回原始目錄
+            os.chdir(original_dir)
+            
+        return result
+
+    except Exception as e:
+        print(f"ERROR in AutoGen processing: {str(e)}")
+        return f"Error analyzing documents: {str(e)}"
+
+
+
+
+
+    
+# === AutoGen 多代理人協作邏輯 ===
+
+        
+def detect_intent_embedding(query, file_names=[]):
+    query_emb = embedding_model.encode(query, normalize_embeddings=True)
+    best_label = None
+    best_score = -1
+    all_phrases = INTENT_LABELS.copy()
+    if file_names:
+        all_phrases["DocQA"] += [name.lower() for name in file_names]
+    for label, examples in all_phrases.items():
+        for example in examples:
+            example_emb = embedding_model.encode(example, normalize_embeddings=True)
+            score = float(query_emb @ example_emb.T)
+            if score > best_score:
+                best_score = score
+                best_label = label
+    return best_label if best_label else "General"
+
+def decide_next(state):
+    query = state.get("query", "")
+    file_names = state.get("file_names", [])
+    label = detect_intent_embedding(query, file_names)
+    return label
+
+# === 定義 Task 物件 ===
+docqa_task = Task(
+    description="Document QA Task: Answer questions based on the uploaded document.",
+    expected_output="Answer from Document QA Agent.",
+    agent=document_qa_agent,
+    input_variables=["query"]
+)
+
+general_task = Task(
+    description="General Chat Task: Answer general queries.",
+    expected_output="Answer from General Agent.",
+    agent=general_agent,
+    input_variables=["query"]
+)
+
+summariser_task = Task(
+    description="Summarisation Task: Summarise document content.",
+    expected_output="Summary output.",
+    agent=summarizer_agent,  # 注意此處名稱須與定義一致（使用字母 z）
+    input_variables=["query"]
+)
+
+search_task = Task(
+    description="Search Task: Retrieve information from the web.",
+    expected_output="Answer from Search Agent.",
+    agent=search_agent,
+    input_variables=["query"]
+)
+
+# === LangGraph 節點函數 ===
+
+def general_run(state):
+    """改用直接 LLM 回答取代 General Agent"""
+    try:
+        prompt = f"""You are a helpful AI assistant. Please answer the following question:
+        {state["query"]}
+        
+        Provide a clear and informative answer."""
+        
+        response = llm_gpt4.invoke(prompt)
+        answer = response.content if hasattr(response, 'content') else str(response)
+        return {"answer": answer}
+    except Exception as e:
+        print(f"ERROR in general_run: {str(e)}")
+        return {"answer": "I apologize, but I'm having trouble processing your request."}
+
+
+def docqa_run(state):
+    """文件問答處理"""
+    try:
+        # 如果有檢索器，使用檢索器
+        if "retriever" in state:
+            relevant_docs = state["retriever"].get_relevant_documents(state["query"])
+            context = "\n".join(d.page_content for d in relevant_docs)
+        else:
+            context = "\n".join(state["docs"])
+            
+        prompt = f"""Based on the following context, please answer the question:
+        Question: {state["query"]}
+        
+        Context:
+        {context[:3000]}
+        
+        Provide a detailed and accurate answer based on the context."""
+        
+        response = llm_gpt4.invoke(prompt)
+        return {"answer": response.content if hasattr(response, 'content') else str(response)}
+    except Exception as e:
+        print(f"ERROR in docqa_run: {str(e)}")
+        return general_run(state)
+
+
+def summariser_run(state):
+    """文件摘要處理"""
+    try:
+        context = "\n".join(state["docs"])
+        prompt = f"""Please provide a comprehensive summary of the following document:
+        {context[:3000]}
+        
+        Focus on:
+        1. Main topics and key points
+        2. Important findings or conclusions
+        3. Significant details"""
+        
+        response = llm_gpt4.invoke(prompt)
+        return {"summary": response.content if hasattr(response, 'content') else str(response)}
+    except Exception as e:
+        print(f"ERROR in summariser_run: {str(e)}")
+        return {"summary": "Error generating summary."}
+
+# === LangGraph 定義 ===
+def build_langgraph_pipeline():
+    graph = StateGraph(dict)
+    graph.add_node("Router", lambda state: state)  # Router 僅傳遞狀態
+    graph.add_node("DocQA", docqa_run)
+    graph.add_node("General", general_run)
+    graph.add_node("Summarise", summariser_run)
+    graph.set_entry_point("Router")
+    graph.add_conditional_edges("Router", decide_next, {
+        "DocQA": "DocQA",
+        "General": "General",
+        "Summarise": "Summarise",
+    })
+    graph.set_finish_point("DocQA")
+    graph.set_finish_point("General")
+    graph.set_finish_point("Summarise")
+    return graph.compile()
+
+from tempfile import mkdtemp
+
+def get_file_path_tab6(file):
+    if isinstance(file, str):
+        print("DEBUG: File is a string:", file)
+        if os.path.exists(file):
+            print("DEBUG: File exists:", file)
+            return file
+        else:
+            print("DEBUG: File does not exist:", file)
+            return None
+    elif isinstance(file, dict):
+        print("DEBUG: File is a dict:", file)
+        data = file.get("data")
+        name = file.get("name")
+        print("DEBUG: Data:", data, "Name:", name)
+        if data:
+            if isinstance(data, str) and os.path.exists(data):
+                print("DEBUG: Data is a valid file path:", data)
+                return data
+            else:
+                temp_dir = mkdtemp()
+                file_path = os.path.join(temp_dir, name if name else "uploaded_file")
+                print("DEBUG: Writing data to temporary file:", file_path)
+                with open(file_path, "wb") as f:
+                    if isinstance(data, str):
+                        f.write(data.encode("utf-8"))
+                    else:
+                        f.write(data)
+                if os.path.exists(file_path):
+                    print("DEBUG: Temporary file created:", file_path)
+                    return file_path
+                else:
+                    print("ERROR: Temporary file not created:", file_path)
+                    return None
+        else:
+            print("DEBUG: No data in dict, returning None")
+            return None
+    elif hasattr(file, "save"):
+        print("DEBUG: File has save attribute")
+        temp_dir = mkdtemp()
+        file_path = os.path.join(temp_dir, file.name)
+        file.save(file_path)
+        if os.path.exists(file_path):
+            print("DEBUG: File saved to:", file_path)
+            return file_path
+        else:
+            print("ERROR: File not saved properly:", file_path)
+            return None
+    else:
+        print("DEBUG: File type unrecognized")
+        if hasattr(file, "name"):
+            if os.path.exists(file.name):
+                return file.name
+        return None
+        
+def langgraph_tab6_main(query: str, file=None):
+    try:
+        print(f"DEBUG: Starting processing with query: {query}")
+        
+        # 如果沒有文件，直接使用 general_run
+        if not file:
+            return general_run({"query": query})["answer"]
+        
+        # 處理文件列表
+        files = file if isinstance(file, list) else [file]
+        all_docs = []
+        file_names = []
+        docs_by_file = []
+        
+        # 處理上傳的文件
+        for f in files:
+            try:
+                path = get_file_path_tab6(f)
+                if not path:
+                    continue
+                
+                file_names.append(os.path.basename(path))
+                
+                # 根據文件類型選擇加載器
+                if path.lower().endswith('.pdf'):
+                    loader = PyPDFLoader(path)
+                elif path.lower().endswith('.docx'):
+                    loader = UnstructuredWordDocumentLoader(path)
+                else:
+                    loader = TextLoader(path)
+                
+                docs = loader.load()
+                if docs:
+                    text = "\n".join(doc.page_content for doc in docs if hasattr(doc, 'page_content'))
+                    docs_by_file.append(text)
+                    all_docs.extend(docs)
+            except Exception as e:
+                print(f"ERROR processing file: {str(e)}")
+                continue
+
+        if not docs_by_file:
+            return general_run({"query": query})["answer"]
+
+        # 建立檢索器
+        try:
+            chunks = RecursiveCharacterTextSplitter(
+                chunk_size=500,
+                chunk_overlap=50
+            ).split_documents(all_docs)
+            
+            db = FAISS.from_documents(chunks, embeddings)
+            retriever = db.as_retriever(search_kwargs={"k": 5})
+            
+            global session_retriever, session_qa_chain
+            session_retriever = retriever
+            session_qa_chain = ConversationalRetrievalChain.from_llm(
+                llm=llm_gpt4,
+                retriever=retriever,
+                memory=ConversationBufferMemory(
+                    memory_key="chat_history",
+                    return_messages=True
+                ),
+            )
+        except Exception as e:
+            print(f"ERROR setting up retriever: {str(e)}")
+            retriever = None
+
+        # 檢測是否為多文件查詢
+        # 檢測是否為多文件查詢或複雜查詢
+        if len(docs_by_file) > 1 or "compare" in query.lower() or "relation" in query.lower():
+            return autogen_multi_document_analysis(query, docs_by_file, file_names)
+            
+        # 使用 LangGraph 處理單文件查詢
+        state = {
+            "query": query,
+            "file_names": file_names,
+            "docs": docs_by_file,
+            "retriever": retriever
+        }
+        
+        # 根據查詢意圖選擇處理方式
+        intent = detect_intent_embedding(query, file_names)
+        if intent == "Summarise":
+            return summariser_run(state)["summary"]
+        elif intent == "DocQA":
+            return docqa_run(state)["answer"]
+        else:
+            return general_run(state)["answer"]
+        
+    except Exception as e:
+        print(f"ERROR in main function: {str(e)}")
+        return f"I apologize, but I encountered an error: {str(e)}"
+
+
+
+
+
 # Gradio Interface Settings 
 demo_description = """
 **Context**:
@@ -1167,14 +1800,52 @@ This demo presents a GPT-style Multi-Agent AI Assistant, built with **LangChain,
 
 Feel free to upload a document and ask related questions, or just type a question directly—no file upload required. *Note: CSV file analysis and auto visualisation is coming soon.*
 """
+demo_description6 = """
+**Context**:
+This is a **smart multi-document reasoning assistant**, powered by **LangGraph**, **CrewAI**, and **AutoGen**.
+Upload zero to multiple files and ask anything — the system will uses **embedding-based intent detection** to decide whether to summarise, extract, compare, or analyse relationships.
+
+For complex multi-file tasks, it triggers a **collaborative AutoGen team** to deeply reason across documents and generate contextual, evidence-based answers.
+
+**Supported Features**:
+- 📄 Multi-document support (PDF, DOCX, TXT)
+- 🔍 Embedding-based intent detection and semantic routing
+- 🤖 Agents: Summariser, QA Agent, General Agent, Search Agent
+- 🔀 Orchestrated by LangGraph + AutoGen (fallbacks + task handoff)
+- 🧠 AutoGen multi-agent collaboration for cross-file reasoning
+- 🌐 Online search fallback if all the other agent can't handle tasks
+
+**Sample Questions**:
+1. Who are you? | What is GPT4? *(→ General Chat Agent)*
+2. Summarise the document/file/your_doc_name. *(→ Summarisation Agent)*
+3. What is LangChain used for? | What are the latest trends in AI startups in 2025? | Tell me the most recent breakthrough in quantum computing. *(→ Online Rag Agent)*
+4. What's the title in the document? | What are the key ideas mentioned in this file? *(→ RAG QA Agent)*
+5. Compare the proposals in DocA and DocB. | Summarise all files. | Is DocA one of the project in the DocB or DocC. | Which argument is stronger across these files? | Do these documents mention similar policies? | What's the difference between the files? *(→ AutoGen)*
+6. What is LangChain used for? | What are the latest trends in AI startups in 2025? | Tell me the most recent breakthrough in quantum computing. *(→ Online Rag Agent)*
+
+> Built for users who need clear, explainable, and context-aware answers — whether you’re working on documents in law, finance, research, or tech.
+"""
+
+
 
 demo = gr.TabbedInterface(
     interface_list=[
+        gr.Interface(
+            fn=langgraph_tab6_main,
+            inputs=[
+                gr.Textbox(label="Ask anything"),
+                gr.File(label="Upload one or more files", file_types=[".pdf", ".txt", ".docx"], file_count="multiple")
+            ],
+            outputs="text",
+            title="Smart Multi-Doc QA (LangGraph + AutoGen)",
+            allow_flagging="never",
+            description=demo_description6
+        ),
         gr.Interface(
             fn=multi_agent_chat_advanced,
             inputs=[
                 gr.Textbox(label="Enter your query"),
-                gr.File(label="Upload file (CSV, PDF, TXT, DOCX)", file_count="single")
+                gr.File(label="Upload file (CSV, PDF, TXT, DOCX)", file_types=[".pdf", ".txt", ".docx"], file_count="single")
             ],
             outputs="text",
             title="Multi-Agent AI Assistant",
@@ -1183,7 +1854,7 @@ demo = gr.TabbedInterface(
         ),
         gr.Interface(
             fn=document_summarize,
-            inputs=[gr.File(label="Upload PDF, TXT, or DOCX")],
+            inputs=[gr.File(label="Upload PDF, TXT, or DOCX", file_types=[".pdf", ".txt", ".docx"])],
             outputs="text",
             title="Document Summarisation",
             allow_flagging="never",
@@ -1191,7 +1862,7 @@ demo = gr.TabbedInterface(
         ),
         gr.Interface(
             fn=upload_and_chat,
-            inputs=[gr.File(label="Upload PDF, TXT, or DOCX"), gr.Textbox(label="Ask a question")],
+            inputs=[gr.File(label="Upload PDF, TXT, or DOCX", file_types=[".pdf", ".txt", ".docx"]), gr.Textbox(label="Ask a question")],
             outputs="text",
             title="Your Docs Q&A (Upload + GPT-4 RAG)",
             allow_flagging="never",
@@ -1213,13 +1884,16 @@ demo = gr.TabbedInterface(
             allow_flagging="never",
             description=demo_description
         ),
+        
     ],
     tab_names=[
+        "Multi-Doc QA",
         "Multi-Agent AI Assistant",
         "Document Summarisation",
         "Your Docs Q&A (Upload + GPT-4 RAG)",
         "Biden Q&A (GPT-4 RAG)",
-        "Biden Q&A (LLaMA RAG)"
+        "Biden Q&A (LLaMA RAG)",
+        
     ],
     title="Smart RAG + Multi-Agent Assistant (with Web + Document AI)"
 )