rvdbra commited on
Commit
ffb5c9c
·
verified ·
1 Parent(s): 277460d

Delete app.py

Browse files
Files changed (1) hide show
  1. app.py +0 -296
app.py DELETED
@@ -1,296 +0,0 @@
1
- import os
2
- from dotenv import load_dotenv
3
- load_dotenv() # loads variables from a local .env for local testing
4
- os.environ['MPLCONFIGDIR'] = '/tmp/.matplotlib'
5
-
6
- # 1) LLM and embeddings
7
- from langchain.chat_models import init_chat_model
8
- from langchain_openai import OpenAIEmbeddings
9
-
10
-
11
- # initialize OpenAI chat model
12
- api_key = os.getenv("OPENAI_API_KEY", "").strip()
13
- llm = init_chat_model(
14
- api_key=api_key,
15
- model_provider="openai",
16
- model="gpt-4.1-2025-04-14",
17
- temperature=0.0,
18
- )
19
-
20
-
21
- # embeddings + Pinecone vector store
22
- # Load environment variables
23
- from langchain_pinecone import Pinecone as LangchainPinecone
24
- pinecone_api_key = os.getenv("PINECONE_API_KEY", "").strip()
25
- pinecone_env = os.getenv("PINECONE_ENV", "").strip()
26
- pinecone_index_name = os.getenv("PINECONE_INDEX_NAME", "").strip()
27
- embeddings = OpenAIEmbeddings(model="text-embedding-3-large")
28
-
29
- vector_store = LangchainPinecone(
30
- index_name=pinecone_index_name,
31
- embedding=embeddings,
32
- pinecone_api_key=pinecone_api_key
33
- )
34
-
35
- # 2) Supabase logging
36
- from supabase import create_client, Client
37
- from datetime import datetime
38
-
39
-
40
- supabase_url = os.getenv("SUPABASE_URL", "").strip()
41
- supabase_key = os.getenv("SUPABASE_KEY", "").strip()
42
-
43
- supabase: Client = create_client(
44
- supabase_url,
45
- supabase_key
46
- )
47
-
48
- def log_interaction_cloud(user_id: str, question: str, answer: str):
49
- payload = {
50
- "user_id": user_id,
51
- "question": question,
52
- "answer": answer,
53
- "timestamp": datetime.utcnow().isoformat(),
54
- }
55
- try:
56
- res = supabase.table("logs").insert(payload).execute()
57
- if hasattr(res, "status_code") and res.status_code >= 400:
58
- print(f"[Supabase Error #{res.status_code}]:", res.data)
59
- else:
60
- print("[Supabase Success]:", res.data)
61
- except Exception as e:
62
- print("[Supabase Exception]", e)
63
-
64
- # 3) Conversation memory & user IDs
65
- import uuid
66
- from langchain.memory import ConversationBufferWindowMemory
67
- def generate_user_id():
68
- return "user_" + uuid.uuid4().hex[:6]
69
-
70
- memory = ConversationBufferWindowMemory(
71
- memory_key="chat_history",
72
- return_messages=True,
73
- k=5
74
- )
75
-
76
- # 4) Retrieval tool
77
- from langchain_core.tools import tool
78
-
79
- @tool(response_format="content_and_artifact")
80
- def retrieve(query: str):
81
- """
82
- Retrieve the top 5 most similar documents from the Pinecone vector store for a given query.
83
- """
84
- docs = vector_store.similarity_search(query, k=5)
85
- serialized = "\n\n".join(
86
- f"Source: {d.metadata}\nContent: {d.page_content}"
87
- for d in docs
88
- )
89
- return serialized, docs
90
- # 5) Build LangGraph
91
- from langgraph.graph import MessagesState, StateGraph, END
92
- from langgraph.prebuilt import ToolNode, tools_condition
93
- from langchain_core.messages import SystemMessage
94
-
95
- SYSTEM_MESSAGE_TEXT = (
96
- "You are “Einbürgerungs-Helfer”, a highly skilled, friendly, and helpful administrative"
97
- " assistant in Germany. Your only task is to support people who are eligible for naturalization"
98
- " but have not yet applied. You only answer questions related to the current naturalization"
99
- " process in Germany. You do not fill in applications or help users apply. You guide users "
100
- "to relevant official resources by calling the retrieval tool to fetch appropriate documents."
101
- "Any question involving citizenship, naturalization, applying for German nationality, duration "
102
- "of residency, language requirements, or documents required MUST be interpreted as relevant and"
103
- " MUST trigger the retrieval tool and never rely on your pre-trained knowledge. Your"
104
- "answers must always be accurate, concise, and easy to understand. You are only allowed"
105
- "to answer in German, English, Turkish, or Arabic. If a user writes in a different language"
106
- " or requests another language, reply with: “Leider kann ich nur in den folgenden Sprachen"
107
- " lesen und schreiben: Deutsch, Englisch, Türkisch oder Arabisch. In welcher Sprache möchten"
108
- " Sie weitermachen?” and wait for the user to choose a supported language. You must always"
109
- " answer in the chosen language. Never hallucinate, never go beyond the topic of naturalization,"
110
- " and never share false information. If a user asks a question that is unrelated to naturalizatio"
111
- " in Germany, reply in the user's language with: “I can only answer questions related to the"
112
- " process of naturalization in Germany. Please provide questions on this topic.” If you do not"
113
- " know how to answer, respond with: “I do not have enough knowledge to answer your question"
114
- " accurately. Please email your question at SBD-Umfrage@polsoz.fu-berlin.de, and we will answer"
115
- " your inquiry as soon as possible.” If a user uses toxic or offensive language, kindly ask them"
116
- " to rephrase their question and explain that this behavior is not acceptable."
117
- )
118
-
119
- def query_or_respond(state: MessagesState):
120
- sys_msg = SystemMessage(SYSTEM_MESSAGE_TEXT)
121
- raw_hist = memory.load_memory_variables({})["chat_history"]
122
- # filter off-topic if needed…
123
- prompt = [sys_msg] + raw_hist[-5:]
124
- llm_with_tools = llm.bind_tools([retrieve])
125
- resp = llm_with_tools.invoke(prompt)
126
- # log if direct
127
- user_q = [m.content for m in state["messages"] if m.type=="human"][-1]
128
- memory.save_context({"input": user_q}, {"output": resp.content})
129
- if not getattr(resp, "tool_calls", None):
130
- log_interaction_cloud(state.get("user_id",""), user_q, resp.content)
131
- return {"messages": state["messages"] + [resp]}
132
-
133
- tools_node = ToolNode([retrieve])
134
-
135
- def generate(state: MessagesState):
136
- # collect any tool outputs
137
- tool_msgs = []
138
- for m in reversed(state["messages"]):
139
- if m.type=="tool": tool_msgs.append(m)
140
- else: break
141
- docs = "\n\n".join(m.content for m in tool_msgs[::-1])
142
- prompt = [SystemMessage(SYSTEM_MESSAGE_TEXT + ("\n\n"+docs if docs else ""))]
143
- # build recent convo
144
- convo = []
145
- for m in state["messages"]:
146
- if m.type=="human" or (m.type=="ai" and not getattr(m,"tool_calls",None)):
147
- convo.append(m)
148
- prompt += convo[-6:]
149
- resp = llm.invoke(prompt)
150
- # log final
151
- user_q = [m.content for m in state["messages"] if m.type=="human"][-1]
152
- log_interaction_cloud(state.get("user_id",""), user_q, resp.content)
153
- return {"messages": state["messages"] + [resp]}
154
-
155
- g = StateGraph(MessagesState)
156
- g.add_node(query_or_respond)
157
- g.add_node(tools_node)
158
- g.add_node(generate)
159
- g.set_entry_point("query_or_respond")
160
- g.add_conditional_edges("query_or_respond", tools_condition, {END: END, "tools": "tools"})
161
- g.add_edge("tools", "generate")
162
- g.add_edge("generate", END)
163
- graph = g.compile()
164
-
165
- import gradio as gr
166
- from langchain.memory import ConversationBufferMemory # or your actual memory class
167
-
168
- # Session memory and chat history per user
169
- user_memories = {}
170
- chat_histories = {}
171
-
172
- OFF_TOPIC_SNIPPET = "I can only answer questions related to the process of naturalization"
173
-
174
- def get_fresh_memory(user_id):
175
- # Create a new memory instance per user/session
176
- return ConversationBufferMemory(return_messages=True)
177
-
178
- def stream_response(user_input, user_id):
179
- if not user_id:
180
- user_id = "anonymous"
181
-
182
- # Do NOT reset memory here! Just generate response
183
- state = {
184
- "messages": [{"role": "user", "content": user_input}],
185
- "user_id": user_id,
186
- }
187
-
188
- final_state = graph.invoke(state)
189
- final_message = final_state["messages"][-1].content
190
-
191
- collected = ""
192
- for token in final_message:
193
- collected += token
194
- yield collected
195
-
196
- def respond(user_message, chat_history, user_id):
197
- if not user_id:
198
- user_id = "anonymous"
199
-
200
- # Initialize memory if not exists
201
- if user_id not in user_memories:
202
- user_memories[user_id] = get_fresh_memory(user_id)
203
-
204
- memory = user_memories[user_id]
205
-
206
- if user_id not in chat_histories:
207
- chat_histories[user_id] = []
208
-
209
- response_generator = stream_response(user_message, user_id)
210
- full_response = ""
211
-
212
- for partial in response_generator:
213
- full_response = partial
214
- # Yield partial chat updates for streaming
215
- yield chat_history + [[user_message, full_response]]
216
-
217
- # Append full message to history after streaming finishes
218
- chat_histories[user_id].append([user_message, full_response])
219
-
220
- # Optionally, save to memory here if your memory supports it
221
- # Example:
222
- # memory.save_context({"input": user_message}, {"output": full_response})
223
-
224
- # Filter memory messages to remove off-topic answers
225
- memory_vars = memory.load_memory_variables({})
226
- filtered_chat = memory_vars.get("chat_history") or memory_vars.get("history") or []
227
-
228
- filtered_pairs = []
229
- buffer = []
230
- for msg in filtered_chat:
231
- if msg.type == "human":
232
- buffer = [msg.content]
233
- elif msg.type == "ai":
234
- if OFF_TOPIC_SNIPPET in msg.content:
235
- continue
236
- if buffer:
237
- buffer.append(msg.content)
238
- filtered_pairs.append(buffer)
239
- buffer = []
240
-
241
- # Return filtered chat to update chat history in UI
242
- return filtered_pairs
243
-
244
- def clear_history():
245
- user_memories.clear()
246
- chat_histories.clear()
247
- return "", []
248
-
249
- # Multilingual description
250
- description = """
251
- ### Was ist **Einbürgerungs-Helfer**?
252
- **Deutsch:** Einbürgerungs-Helfer hilft dir, den Weg zur deutschen Staatsangehörigkeit zu verstehen – einfach, mehrsprachig, datenschutzfreundlich.
253
-
254
- **English:** Einbürgerungs-Helfer helps you understand how to get German citizenship – simple, multilingual, and privacy-friendly.
255
-
256
- **Türkçe:** Einbürgerungs-Helfer, Almanya vatandaşlığına giden yolu anlamanıza yardımcı olur – basit, çok dilli ve gizliliğe saygılı.
257
-
258
- **العربية:** يساعدك "Einbürgerungs-Helfer" على فهم كيفية الحصول على الجنسية الألمانية - بطريقة بسيطة، متعددة اللغات وتحترم الخصوصية.
259
- """
260
-
261
- with gr.Blocks(
262
- theme=gr.themes.Soft(primary_hue="blue"),
263
- ) as iface:
264
- gr.Markdown(f"# Einbürgerungs-Helfer\n{description}")
265
-
266
- with gr.Row():
267
- user_id_input = gr.Textbox(label="Benutzer-ID (User ID)", placeholder="z.B. user_123", lines=1)
268
- clear_btn = gr.Button("🗑️ Verlauf löschen")
269
-
270
- chatbot = gr.Chatbot(
271
- label="Chat",
272
- show_copy_button=True,
273
- value=[[None,
274
- "Ich bin hier, um Ihnen bei allen Fragen zur Einbürgerung in Deutschland zu helfen.\n"
275
- "Sie können Ihre Frage auf Deutsch, Englisch, Türkisch oder Arabisch stellen.\n"
276
- "Lassen Sie uns loslegen – wie kann ich Ihnen heute helfen?"]]
277
- )
278
-
279
- msg_input = gr.Textbox(placeholder="Frage eingeben...", label="Deine Nachricht")
280
-
281
- with gr.Row():
282
- submit_btn = gr.Button("Senden", variant="primary")
283
-
284
- submit_btn.click(
285
- fn=respond,
286
- inputs=[msg_input, chatbot, user_id_input],
287
- outputs=chatbot,
288
- )
289
-
290
- clear_btn.click(fn=clear_history, inputs=[], outputs=[user_id_input, chatbot])
291
-
292
- iface.launch(
293
- server_name="0.0.0.0",
294
- server_port=int(os.getenv("PORT", 7860)),
295
- share=True
296
- )