Update app.py

app.py

@@ -1,59 +1,102 @@
 import gradio as gr
-import requests
-import os
+import json
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain_community.document_loaders import WebBaseLoader
+from langchain_community.embeddings import HuggingFaceEmbeddings
+from langchain_community.vectorstores import FAISS
+from langchain.memory import ConversationBufferMemory
+from langchain.chains import ConversationalRetrievalChain
+from langchain_huggingface import HuggingFaceEndpoint
 
-# --- Flowise endpoint ---
-# Nếu bạn chạy Flowise cục bộ: ngrok http 3000
-FLOWISE_API = os.getenv("FLOWISE_API", "https://<your-ngrok-or-render-url>/api/v1/prediction")
+# -----------------------------
+# 1️⃣ Load pipeline JSON (Flowise export)
+# -----------------------------
+PIPELINE_PATH = "Chat1 Chatflow.json"
+with open(PIPELINE_PATH, "r") as f:
+    pipeline = json.load(f)
 
+# Lấy URL trong pipeline (nếu có)
+web_url = None
+for node in pipeline.get("nodes", []):
+    if "url" in node.get("data", {}):
+        web_url = node["data"]["url"]
+
+if not web_url:
+    web_url = "https://recht.nrw.de/lmi/owa/br_text_anzeigen?v_id=10000000000000000654"
+
+# -----------------------------
+# 2️⃣ Load documents (from website)
+# -----------------------------
+print(f"🌐 Lade Text von: {web_url}")
+loader = WebBaseLoader(web_url)
+docs = loader.load()
+
+splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=150)
+docs_split = splitter.split_documents(docs)
+
+# -----------------------------
+# 3️⃣ Embeddings + Vectorstore
+# -----------------------------
+embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
+vectorstore = FAISS.from_documents(docs_split, embedding=embeddings)
+retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
+
+# -----------------------------
+# 4️⃣ LLM + Memory
+# -----------------------------
+llm = HuggingFaceEndpoint(
+    repo_id="mistralai/Mistral-7B-Instruct-v0.2",
+    temperature=0.0,
+    huggingfacehub_api_token=None  # Hugging Face Space đã tự có quyền
+)
+
+memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
+
+# -----------------------------
+# 5️⃣ RAG Chain
+# -----------------------------
+qa_chain = ConversationalRetrievalChain.from_llm(
+    llm=llm,
+    retriever=retriever,
+    memory=memory,
+    return_source_documents=True
+)
+
+# -----------------------------
+# 6️⃣ Chat function
+# -----------------------------
 def chat_with_bot(message, history):
-    """
-    Gửi câu hỏi người dùng đến Flowise API và nhận câu trả lời.
-    """
-    payload = {"question": message}
-    try:
-        response = requests.post(FLOWISE_API, json=payload, timeout=60)
-        data = response.json()
-
-        answer = data.get("text", "Keine Antwort gefunden.")
-        sources = data.get("sourceDocuments", [])
-
-        if sources:
-            refs = "\n\n**Quellen:**\n" + "\n".join([
-                f"- {src.get('metadata', {}).get('source', 'Unbekannt')}"
-                for src in sources
-            ])
-        else:
-            refs = ""
-
-        return answer + refs
-
-    except Exception as e:
-        return f"Fehler: {e}"
+    result = qa_chain({"question": message})
+    answer = result.get("answer", "Keine Antwort gefunden.")
+    sources = result.get("source_documents", [])
+    if sources:
+        refs = "\n\n**Quellen:**\n" + "\n".join([
+            f"- {src.metadata.get('source', 'Unbekannt')}"
+            for src in sources
+        ])
+    else:
+        refs = ""
+    return answer + refs
 
+# -----------------------------
+# 7️⃣ UI
+# -----------------------------
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("# 🤖 Chatbot für Prüfungsrecht")
-    gr.Markdown("Fragen Sie zur Prüfungsordnung oder zum Hochschulgesetz NRW. "
-                "Der Chatbot zitiert direkt aus den Quellen (PDF & Webseite).")
+    gr.Markdown("## 🤖 Chatbot für Prüfungsrecht")
+    gr.Markdown(
+        "Fragen Sie zur Prüfungsordnung oder zum Hochschulgesetz NRW. "
+        "Der Chatbot antwortet auf Basis der aus Flowise exportierten Pipeline "
+        "und zitiert direkt aus der Quelle (recht.nrw.de)."
+    )
 
     chat = gr.ChatInterface(
         fn=chat_with_bot,
-        title="Prüfungsrecht RAG-Chatbot",
+        title="Prüfungsrecht RAG-Chatbot (Serverless)",
         examples=[
             "Was steht in §10 über Wiederholungsprüfungen?",
-            "Welche Regel gilt laut Hochschulgesetz für Prüfungsanspruch?",
+            "Welche Regel gilt laut Hochschulgesetz für Prüfungsanspruch?"
         ]
     )
 
-    gr.HTML("""
-    <div style='display:flex;gap:10px;'>
-        <iframe src="https://recht.nrw.de/lmi/owa/br_text_anzeigen?v_id=10000000000000000654"
-              width="49%" height="400"></iframe>
-        <iframe src="https://huggingface.co/spaces/<your-space-name>/resolve/main/PO_BWI_THK_2022.pdf"
-              width="49%" height="400"></iframe>
-    </div>
-    """)
-
-
 if __name__ == "__main__":
     demo.launch()