Update app.py

app.py

@@ -94,6 +94,9 @@ 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
+
+
 
 try:
     from phoenix.trace.langchain import LangChainInstrumentor
@@ -1238,12 +1241,10 @@ def multi_agent_chat_advanced(query: str, file=None) -> str:
 
 
 # LangGraph 使用的節點函數（會接續你的 Crew Agent）
-# ✅ Tab6 最終設計：General + Search fallback 接在 DocQA 後面，Summarise 平行觸發
-from sentence_transformers import SentenceTransformer
 
 embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
 
-# 用於 LangGraph 路由的意圖標籤
+# Intent Embedding 分類（支援檔名）
 INTENT_LABELS = {
     "DocQA": ["document", "file", "paper", "cb", "proposal"],
     "Summarise": ["summarise", "summary", "abstract", "key points", "要點", "總結", "摘要"],
@@ -1252,8 +1253,9 @@ INTENT_LABELS = {
 
 def detect_intent_embedding(query, file_names=[]):
     query_emb = embedding_model.encode(query, normalize_embeddings=True)
-    best_label, best_score = None, -1
 
+    best_label = None
+    best_score = -1
     all_phrases = INTENT_LABELS.copy()
     if file_names:
         all_phrases["DocQA"] += [name.lower() for name in file_names]
@@ -1263,59 +1265,70 @@ def detect_intent_embedding(query, file_names=[]):
             example_emb = embedding_model.encode(example, normalize_embeddings=True)
             score = float(query_emb @ example_emb.T)
             if score > best_score:
-                best_score, best_label = score, label
-    return best_label if best_label else "DocQA"
+                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", [])
-    return detect_intent_embedding(query, file_names)
+    label = detect_intent_embedding(query, file_names)
+    return label
+
+# ✅ node functions
 
-# ⛓️ LangGraph 節點函數
+def docqa_run(state):
+    result = document_qa_agent.execute_task(docqa_task, inputs={"query": state["query"]})
+    output = result.output.lower()
+    # fallback: no info found → general
+    if any(x in output for x in ["no relevant info", "not found", "no answer"]):
+        return general_run(state)
+    return {"answer": result.output}
 
-def docqa_run(state: dict) -> dict:
-    result = document_qa_agent.execute_task(router_task, inputs={"query": state["query"]})
+def general_run(state):
+    result = general_agent.execute_task(general_task, inputs={"query": state["query"]})
     output = result.output.lower()
-    if any(x in output for x in ["no relevant", "can't find", "not in document"]):
-        result = general_agent.execute_task(router_task, inputs={"query": state["query"]})
-        output = result.output.lower()
-        if any(x in output for x in ["i don't know", "no idea", "not sure", "can't answer", "no info"]):
-            result = search_agent.execute_task(router_task, inputs={"query": state["query"]})
-    return {"query": state["query"], "answer": result.output}
+    if any(x in output for x in ["i don't know", "no idea", "not sure", "can't answer"]):
+        result = search_agent.execute_task(search_task, inputs={"query": state["query"]})
+    return {"answer": result.output}
 
-def summarizer_run(state: dict) -> dict:
-    result = summarizer_agent.execute_task(router_task, inputs={"query": state["query"]})
-    return {"query": state["query"], "summary": result.output}
+def summariser_run(state):
+    result = summariser_agent.execute_task(summariser_task, inputs={"query": state["query"]})
+    return {"summary": result.output}
 
-# LangGraph 設計
+# ✅ LangGraph 定義
 
-def build_langgraph_gpt_like():
+def build_langgraph_pipeline():
     graph = StateGraph(dict)
 
+    graph.add_node("Router", lambda state: state)  # router 只傳入狀態即可
     graph.add_node("DocQA", docqa_run)
-    graph.add_node("Summarise", summarizer_run)
+    graph.add_node("General", general_run)
+    graph.add_node("Summarise", summariser_run)
 
-    graph.set_entry_point("DocQA")
-    graph.add_conditional_edges("DocQA", decide_next, {
+    graph.set_entry_point("Router")
+    graph.add_conditional_edges("Router", decide_next, {
+        "DocQA": "DocQA",
+        "General": "General",
         "Summarise": "Summarise",
-        "DocQA": "DocQA"  # 防止空跳轉
     })
 
     graph.set_finish_point("DocQA")
+    graph.set_finish_point("General")
     graph.set_finish_point("Summarise")
+
     return graph.compile()
 
-# 主入口函數
+# ✅ 主執行函數
 
-def langgraph_tab6_main(query: str, file=None) -> str:
+def langgraph_tab6_main(query: str, file=None):
     try:
         files = file if isinstance(file, list) else [file] if file else []
         all_docs, file_names = [], []
 
         for f in files:
             path = get_file_path(f)
-            if not path:
-                continue
+            if not path: continue
             file_names.append(os.path.basename(path))
             if path.lower().endswith(".pdf"):
                 loader = PyPDFLoader(path)
@@ -1325,29 +1338,29 @@ def langgraph_tab6_main(query: str, file=None) -> str:
                 loader = TextLoader(path)
             all_docs.extend(loader.load())
 
-        # 即使沒檔案也進入 QA，但內容會讓 QA 回答不了 → fallback 到 general → fallback 到 search
-        if all_docs:
+        if not all_docs:
+            retriever = None  # 空 retriever
+        else:
             chunks = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50).split_documents(all_docs)
             db = FAISS.from_documents(chunks, embeddings)
             retriever = db.as_retriever()
-            # 替換 DocQA agent retriever
-            document_qa_agent.tools[0].retriever = retriever
 
-        graph = build_langgraph_gpt_like()
-        result = graph.invoke({"query": query, "file_names": file_names})
+        # 設定 retriever 到 global Agent，如果你需要可傳給 Agent
+        # 可選：document_qa_agent.retriever = retriever
+
+        graph = build_langgraph_pipeline()
+        state = {"query": query, "file_names": file_names}
+        result = graph.invoke(state)
 
-        if "answer" in result and "summary" in result:
-            return f"\nAnswer: {result['answer']}\n\nSummary: {result['summary']}"
-        elif "answer" in result:
+        if "answer" in result:
             return result["answer"]
-        elif "summary" in result:
+        if "summary" in result:
             return result["summary"]
-        else:
-            return "No result."
-
+        return "No answer."
     except Exception as e:
         return f"[Tab6 Error] {e}"
 
+
     
 
 # Gradio Interface Settings