rvdbra commited on
Commit
680983c
·
verified ·
1 Parent(s): f6e36df

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +168 -98
app.py CHANGED
@@ -1,146 +1,216 @@
1
  import os
2
  import uuid
3
  from datetime import datetime
 
4
  from dotenv import load_dotenv
5
 
6
  import gradio as gr
7
- import pinecone
 
 
 
8
  from supabase import create_client, Client
9
- from langchain.chat_models import ChatOpenAI
10
- from langchain.embeddings.openai import OpenAIEmbeddings
11
- from langchain.vectorstores import Pinecone as LangchainPinecone
12
- from langchain.memory import ConversationBufferMemory
13
- from langchain.chains import ConversationalRetrievalChain
14
 
15
- # ─── Load environment variables ────────────────────────────────────────────────
16
  load_dotenv()
17
- OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
18
- PINECONE_API_KEY = os.getenv("PINECONE_API_KEY")
19
- PINECONE_ENV = os.getenv("PINECONE_ENVIRONMENT", "us-east1-gcp")
20
- PINECONE_INDEX = os.getenv("PINECONE_INDEX", "natstudy")
21
- SUPABASE_URL = os.getenv("SUPABASE_URL")
22
- SUPABASE_KEY = os.getenv("SUPABASE_KEY")
23
-
24
- # ─── Initialize LLM ────────────────────────────────────────────────────────────
25
- llm = ChatOpenAI(
26
- model_name="gpt-4",
27
- temperature=0.0,
28
- openai_api_key=OPENAI_API_KEY
29
- )
30
-
31
- # ─── Initialize Pinecone + VectorStore ────────────────────────────────────────
32
- pinecone.init(api_key=PINECONE_API_KEY, environment=PINECONE_ENV)
33
- embeddings = OpenAIEmbeddings(openai_api_key=OPENAI_API_KEY)
34
- vectorstore = LangchainPinecone.from_existing_index(
35
  index_name=PINECONE_INDEX,
36
- embedding=embeddings
 
37
  )
38
 
39
- # ─── Initialize Supabase client ───────────────────────────────────────────────
40
  supabase: Client = create_client(SUPABASE_URL, SUPABASE_KEY)
41
 
42
  def log_interaction_cloud(user_id: str, question: str, answer: str):
43
- """Log each Q&A to your Supabase `logs` table."""
44
  payload = {
45
- "user_id": user_id,
46
  "question": question,
47
- "answer": answer,
48
  "timestamp": datetime.utcnow().isoformat(),
49
  }
50
  try:
51
- res = supabase.from_("logs").insert(payload).execute()
52
- print("Logged to Supabase:", res)
53
  except Exception as e:
54
- print("Supabase logging error:", e)
55
-
56
- def generate_user_id() -> str:
 
 
 
 
 
57
  return "user_" + uuid.uuid4().hex[:6]
58
 
59
- # ─── Per-user memory store ─────────────────────────────────────────────────────
60
- user_memories = {}
 
 
 
61
 
62
- def get_memory(user_id: str) -> ConversationBufferMemory:
63
- if user_id not in user_memories:
64
- user_memories[user_id] = ConversationBufferMemory(
65
- memory_key="chat_history",
66
- return_messages=True,
67
- k=5
68
- )
69
- return user_memories[user_id]
 
70
 
71
- # ─── A friendly greeting to start the conversation ─────────────────────────────
72
- GREETING = (
73
- "Ich bin hier, um Ihnen bei allen Fragen zur Einbürgerung in Deutschland zu helfen.\n"
74
- "Sie können Ihre Frage auf Deutsch, Englisch, Türkisch oder Arabisch stellen.\n"
75
- "Lassen Sie uns loslegen – wie kann ich Ihnen heute helfen?"
76
  )
77
 
78
- # ─── The main Gradio callback ───────────────────────────────────────────────────
79
- def respond(user_message, chat_history, user_id):
80
- # assign a new user_id if none provided
81
- if not user_id:
82
- user_id = generate_user_id()
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
83
 
84
- # initialize chat history with greeting if this is the first turn
85
- if not chat_history:
86
- chat_history = [[None, GREETING]]
 
 
 
 
 
 
 
87
 
88
- # fetch (or create) the user's memory buffer
89
- memory = get_memory(user_id)
 
 
 
 
 
90
 
91
- # build a conversational retrieval chain
92
- chain = ConversationalRetrievalChain.from_llm(
93
- llm=llm,
94
- retriever=vectorstore.as_retriever(),
95
- memory=memory,
96
- return_source_documents=False,
97
- )
98
 
99
- # run the chain
100
- result = chain({"question": user_message})
101
- answer = result["answer"]
 
102
 
103
- # log to Supabase
104
- log_interaction_cloud(user_id, user_message, answer)
 
 
 
105
 
106
- # append to our Gradio history
107
- chat_history.append([user_message, answer])
108
- return chat_history, user_id
109
 
110
- # ─── Clear button resets user_id & chat history ────────────────────────────────
111
  def clear_history():
112
  return "", []
113
 
114
- # ─── Gradio UI ─────────────────────────────────────────────────────────────────
115
- with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue")) as iface:
116
- gr.Markdown(
117
- "# Einbürgerungs-Helfer\n"
118
- "Einbürgerungs-Helfer hilft dir, den Weg zur deutschen Staatsangehörigkeit zu verstehen – "
119
- "einfach, mehrsprachig und datenschutzfreundlich."
120
- )
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
121
 
122
  with gr.Row():
123
- user_id_input = gr.Textbox(
124
- label="Benutzer-ID (User ID)",
125
- placeholder="z.B. user_123",
126
- lines=1
127
- )
128
  clear_btn = gr.Button("🗑️ Verlauf löschen")
129
 
130
- chatbot = gr.Chatbot(label="Chat", show_copy_button=True)
 
 
 
 
 
 
 
 
131
  msg_input = gr.Textbox(placeholder="Frage eingeben...", label="Deine Nachricht")
132
- submit_btn = gr.Button("Senden", variant="primary")
 
 
133
 
134
  submit_btn.click(
135
  fn=respond,
136
  inputs=[msg_input, chatbot, user_id_input],
137
- outputs=[chatbot, user_id_input]
138
- )
139
- clear_btn.click(
140
- fn=clear_history,
141
- inputs=[],
142
- outputs=[user_id_input, chatbot]
143
  )
144
 
145
- if __name__ == "__main__":
146
- iface.launch(server_name="0.0.0.0", server_port=7860)
 
 
 
1
  import os
2
  import uuid
3
  from datetime import datetime
4
+ from typing_extensions import TypedDict
5
  from dotenv import load_dotenv
6
 
7
  import gradio as gr
8
+ from langchain_openai import OpenAIEmbeddings
9
+ from langchain.chat_models import init_chat_model
10
+ from langchain_community.vectorstores import Pinecone
11
+ from pinecone import Pinecone
12
  from supabase import create_client, Client
13
+ from langchain.memory import ConversationBufferWindowMemory
14
+ from langchain_core.messages import SystemMessage
15
+ from langchain_core.tools import tool
16
+ from langgraph.graph import StateGraph, END
17
+ from langgraph.prebuilt import ToolNode, tools_condition
18
 
19
+ # Load environment variables
20
  load_dotenv()
21
+ SUPABASE_URL = os.environ.get("SUPABASE_URL")
22
+ SUPABASE_KEY = os.environ.get("SUPABASE_KEY")
23
+ OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY")
24
+ LANGSMITH_API_KEY = os.environ.get("LANGSMITH_API_KEY")
25
+ PINECONE_API_KEY = os.environ.get("PINECONE_API_KEY") or "fallback_key"
26
+ PINECONE_INDEX = os.environ.get("PINECONE_INDEX") or "natstudy"
27
+
28
+ # Model and vector store setup
29
+ llm = init_chat_model("gpt-4.1-2025-04-14", model_provider="openai", temperature=0.0)
30
+ embeddings = OpenAIEmbeddings(model="text-embedding-3-large")
31
+ pc = Pinecone(api_key=PINECONE_API_KEY)
32
+ index = pc.Index(PINECONE_INDEX)
33
+ vector_store = Pinecone(
34
+ embedding=embeddings,
 
 
 
 
35
  index_name=PINECONE_INDEX,
36
+ api_key=PINECONE_API_KEY,
37
+ environment=os.environ.get("PINECONE_ENVIRONMENT", "us-east1-gcp")
38
  )
39
 
40
+ # Supabase
41
  supabase: Client = create_client(SUPABASE_URL, SUPABASE_KEY)
42
 
43
  def log_interaction_cloud(user_id: str, question: str, answer: str):
 
44
  payload = {
45
+ "user_id": user_id,
46
  "question": question,
47
+ "answer": answer,
48
  "timestamp": datetime.utcnow().isoformat(),
49
  }
50
  try:
51
+ res = supabase.table("logs").insert(payload).execute()
 
52
  except Exception as e:
53
+ print("[Supabase Exception]", e)
54
+ return
55
+ if hasattr(res, "status_code") and res.status_code >= 400:
56
+ print(f"[Supabase Error #{res.status_code}]:", res.data)
57
+ else:
58
+ print("[Supabase Success]:", res.data)
59
+
60
+ def generate_user_id():
61
  return "user_" + uuid.uuid4().hex[:6]
62
 
63
+ memory = ConversationBufferWindowMemory(
64
+ memory_key="chat_history",
65
+ return_messages=True,
66
+ k=5
67
+ )
68
 
69
+ @tool(response_format="content_and_artifact")
70
+ def retrieve(query: str):
71
+ """Retrieve information related to the last query."""
72
+ retrieved_docs = vector_store.similarity_search(query, k=5)
73
+ serialized = "\n\n".join(
74
+ (f"Source: {doc.metadata}\nContent: {doc.page_content}")
75
+ for doc in retrieved_docs
76
+ )
77
+ return serialized, retrieved_docs
78
 
79
+ SYSTEM_MESSAGE_TEXT = (
80
+ "You are “Einbürgerungs-Helfer”, a highly skilled, friendly, and helpful administrative "
81
+ "assistant in Germany. Your only task is to support people who are eligible for naturalization "
82
+ "but have not yet applied. [...]" # Full instructions remain unchanged for brevity
 
83
  )
84
 
85
+ class MyState(TypedDict):
86
+ messages: list
87
+ user_id: str
88
+
89
+ def query_or_respond(state: MyState):
90
+ system_message = SystemMessage(SYSTEM_MESSAGE_TEXT)
91
+ raw_history = memory.load_memory_variables({})["chat_history"]
92
+ filtered_history = [msg for msg in raw_history if not (
93
+ msg.type == "ai" and "I can only answer questions related" in msg.content)]
94
+ prompt = [system_message] + filtered_history[-5:]
95
+ llm_with_tools = llm.bind_tools([retrieve])
96
+ response = llm_with_tools.invoke(prompt)
97
+ user_id = state.get("user_id", "unknown_user")
98
+ user_question = [m["content"] for m in state["messages"] if m["type"] == "human"][-1]
99
+ memory.save_context({"input": user_question}, {"output": response.content})
100
+ if not getattr(response, "tool_calls", None):
101
+ log_interaction_cloud(user_id, user_question, response.content)
102
+ return {"messages": state["messages"] + [response], "user_id": user_id}
103
+
104
+ def generate(state: MyState):
105
+ last_message = state["messages"][-1]
106
+ user_id = state.get("user_id", "unknown_user")
107
+ user_question = [m["content"] for m in state["messages"] if m["type"] == "human"][-1]
108
+ answer_text = getattr(last_message, "content", str(last_message))
109
+ log_interaction_cloud(user_id, user_question, answer_text)
110
+ return {"messages": state["messages"], "user_id": user_id}
111
+
112
+ # LangGraph setup
113
+ workflow = StateGraph(MyState)
114
+ workflow.add_node("query_or_respond", query_or_respond)
115
+ workflow.add_node("tool", ToolNode([retrieve]))
116
+ workflow.add_node("generate", generate)
117
+ workflow.add_conditional_edges("query_or_respond", tools_condition, {
118
+ "tool": "tool",
119
+ "generate": "generate"
120
+ })
121
+ workflow.add_edge("tool", "generate")
122
+ workflow.add_edge("generate", END)
123
+ workflow.set_entry_point("query_or_respond")
124
+ graph = workflow.compile()
125
+
126
+ # Gradio UI
127
+ import gradio as gr
128
 
129
+ # Global chat history (for demo purposes)
130
+ chat_histories = {}
131
+
132
+ def stream_response(user_input, user_id):
133
+ if not user_id:
134
+ user_id = "anonymous"
135
+ state = {
136
+ "messages": [{"role": "user", "content": user_input}],
137
+ "user_id": user_id,
138
+ }
139
 
140
+ collected = ""
141
+ for step in graph.stream(state, stream_mode="values"):
142
+ message = step["messages"][-1].content
143
+ # Stream response token by token
144
+ for token in message:
145
+ collected += token
146
+ yield collected
147
 
 
 
 
 
 
 
 
148
 
149
+
150
+ def respond(user_message, chat_history, user_id):
151
+ if user_id not in chat_histories:
152
+ chat_histories[user_id] = []
153
 
154
+ response_generator = stream_response(user_message, user_id)
155
+ full_response = ""
156
+ for partial in response_generator:
157
+ full_response = partial
158
+ yield chat_history + [[user_message, full_response]]
159
 
160
+ chat_histories[user_id].append([user_message, full_response])
 
 
161
 
 
162
  def clear_history():
163
  return "", []
164
 
165
+ # Multilingual description
166
+ description = """
167
+ ### Was ist **Einbürgerungs-Helfer**?
168
+ **Deutsch:** Einbürgerungs-Helfer hilft dir, den Weg zur deutschen Staatsangehörigkeit zu verstehen – einfach, mehrsprachig, datenschutzfreundlich.
169
+
170
+ **English:** Einbürgerungs-Helfer helps you understand how to get German citizenship – simple, multilingual, and privacy-friendly.
171
+
172
+ **Türkçe:** Einbürgerungs-Helfer, Almanya vatandaşlığına giden yolu anlamanıza yardımcı olur – basit, çok dilli ve gizliliğe saygılı.
173
+
174
+ **العربية:** يساعدك "Einbürgerungs-Helfer" على فهم كيفية الحصول على الجنسية الألمانية - بطريقة بسيطة، متعددة اللغات وتحترم الخصوصية.
175
+ """
176
+
177
+ with gr.Blocks(
178
+ theme=gr.themes.Soft(primary_hue="blue"),
179
+ css="""
180
+ @import url('https://fonts.googleapis.com/css2?family=Noto+Sans&display=swap');
181
+
182
+ * {
183
+ font-family: 'Noto Sans', sans-serif !important;
184
+ }
185
+ """
186
+ ) as iface:
187
+ gr.Markdown(f"# Einbürgerungs-Helfer\n{description}")
188
 
189
  with gr.Row():
190
+ user_id_input = gr.Textbox(label="Benutzer-ID (User ID)", placeholder="z.B. user_123", lines=1)
 
 
 
 
191
  clear_btn = gr.Button("🗑️ Verlauf löschen")
192
 
193
+ chatbot = gr.Chatbot(
194
+ label="Chat",
195
+ show_copy_button=True,
196
+ value=[[None,
197
+ "Ich bin hier, um Ihnen bei allen Fragen zur Einbürgerung in Deutschland zu helfen.\n"
198
+ "Sie können Ihre Frage auf Deutsch, Englisch, Türkisch oder Arabisch stellen.\n"
199
+ "Lassen Sie uns loslegen – wie kann ich Ihnen heute helfen?"]]
200
+ )
201
+
202
  msg_input = gr.Textbox(placeholder="Frage eingeben...", label="Deine Nachricht")
203
+
204
+ with gr.Row():
205
+ submit_btn = gr.Button("Senden", variant="primary")
206
 
207
  submit_btn.click(
208
  fn=respond,
209
  inputs=[msg_input, chatbot, user_id_input],
210
+ outputs=chatbot,
 
 
 
 
 
211
  )
212
 
213
+ clear_btn.click(fn=clear_history, inputs=[], outputs=[user_id_input, chatbot])
214
+
215
+ iface.launch(server_name="0.0.0.0", server_port=7860)
216
+