VectorV17 commited on
Commit
29345ee
·
verified ·
1 Parent(s): a02a082

Create app.py

Browse files
Files changed (1) hide show
  1. app.py +120 -298
app.py CHANGED
@@ -1,328 +1,150 @@
 
1
  import gradio as gr
2
  from huggingface_hub import InferenceClient
3
- from datetime import datetime
4
- import re
5
-
6
- # Import NLP and Math Matching frameworks
7
- import nltk
8
- from sklearn.feature_extraction.text import TfidfVectorizer
9
- from sklearn.metrics.pairwise import cosine_similarity
10
-
11
- # Initialize NLTK tokenization assets safely on basic CPU bootup
12
- try:
13
- nltk.download('punkt', quiet=True)
14
- nltk.download('punkt_tab', quiet=True)
15
- except Exception:
16
- pass # Fallback to standard regex splitting if environment is strictly offline
17
-
18
- # Initialize the Hugging Face Serverless client
19
- client = InferenceClient("Qwen/Qwen2.5-7B-Instruct")
20
-
21
- # 1. Collect FAQs related to topics/products into a structured corpus
22
- FAQ_DATA = [
23
- {
24
- "q": "What are your shipping times and policies?",
25
- "a": "Shipping times vary by location. Typically, local retail deliveries take 2-3 business days, while international dropshipping options take 7-14 business days. All tracking codes are sent via email."
26
- },
27
- {
28
- "q": "What is the return policy for products?",
29
- "a": "We offer a flexible 30-day return policy. Items must be unworn, in their original packaging layout, and accompanied by the checkout receipt."
30
- },
31
- {
32
- "q": "What are your service project timelines and quotes?",
33
- "a": "Standard design and consulting project timelines range between 2 to 6 weeks. You can request a personalized pricing quote directly via our booking panel."
34
- },
35
- {
36
- "q": "What are the differences between your software subscription tiers?",
37
- "a": "We offer Free, Pro, and Enterprise tiers. Pro unlocks comprehensive API documentation access, and Enterprise unlocks custom third-party platform integrations."
38
- },
39
- {
40
- "q": "How do I unlock my course completion certificate?",
41
- "a": "Completion certificates unlock automatically on your student profile dashboard as soon as all video modules and the final evaluation quiz are marked 100% complete."
42
- },
43
- {
44
- "q": "What insurance providers do you accept at the clinic?",
45
- "a": "We accept major national medical insurance providers. Please contact our front office coordinator to verify your specific policy layout before scheduling appointments."
46
- },
47
- {
48
- "q": "What are your standard check-in and check-out times?",
49
- "a": "Standard hospitality check-in opens at 3:00 PM, and checkout must be finalized by 11:00 AM. Early check-in requests can be accommodated based on live room availability."
50
- }
51
- ]
52
-
53
- def perform_nlp_cosine_matching(user_query):
54
- """
55
- Preprocesses text using tokenization and applies Cosine Similarity
56
- matching over TF-IDF vectors to find the best matching FAQ.
57
- """
58
- try:
59
- tokens = nltk.word_tokenize(user_query.lower())
60
- cleaned_query = " ".join([w for w in tokens if re.match(r'^\w+$', w)])
61
- except Exception:
62
- cleaned_query = " ".join(re.findall(r'\w+', user_query.lower()))
63
-
64
- if not cleaned_query.strip():
65
- return None, 0.0
66
-
67
- questions_corpus = [faq["q"] for faq in FAQ_DATA]
68
-
69
- vectorizer = TfidfVectorizer(stop_words='english')
70
- tfidf_matrix = vectorizer.fit_transform(questions_corpus)
71
-
72
- query_vector = vectorizer.transform([cleaned_query])
73
- similarity_scores = cosine_similarity(query_vector, tfidf_matrix)[0]
74
-
75
- best_match_idx = similarity_scores.argmax()
76
- highest_score = similarity_scores[best_match_idx]
77
-
78
- return FAQ_DATA[best_match_idx], float(highest_score)
79
-
80
- def get_thematic_datetime():
81
- """Generates precise, real-time updated date and running seconds strings."""
82
- now = datetime.now()
83
- day_int = now.day
84
- if 11 <= day_int <= 13:
85
- suffix = "th"
86
- else:
87
- suffix = {1: "st", 2: "nd", 3: "rd"}.get(day_int % 10, "th")
88
 
89
- formatted_date = f"{now.strftime('%d')}{suffix} {now.strftime('%B %Y')} | Time: {now.strftime('%I:%M:%S %p').lower()}"
90
- return f"<div style='text-align: right; font-weight: 600; font-size: 14px; font-family: monospace; letter-spacing: 0.5px;'>{formatted_date}</div>"
91
-
92
- def extract_text(content):
93
- """Safely extracts plain string text from any Gradio 6 message structure."""
94
- if isinstance(content, str):
95
- return content
96
- if isinstance(content, list):
97
- extracted_parts = []
98
- for block in content:
99
- if isinstance(block, dict) and "text" in block:
100
- extracted_parts.append(block["text"])
101
- elif isinstance(block, str):
102
- extracted_parts.append(block)
103
- return "".join(extracted_parts)
104
- if isinstance(content, dict):
105
- if "text" in content: return content["text"]
106
- if "content" in content: return content["content"]
107
- return str(content)
108
-
109
- def user_side(user_message, history):
110
- """Handles incoming user messages and sanitizes state history format."""
111
  if not user_message.strip():
112
- return "", history
113
-
114
- clean_history = []
115
- for msg in history:
116
- if isinstance(msg, dict):
117
- clean_history.append({"role": msg["role"], "content": extract_text(msg["content"])})
118
- elif hasattr(msg, "role") and hasattr(msg, "content"):
119
- clean_history.append({"role": msg.role, "content": extract_text(msg.content)})
120
-
121
- clean_history.append({"role": "user", "content": user_message})
122
- return "", clean_history
123
-
124
- def bot_side(history, system_message, max_tokens, temperature, top_p):
125
- """Streams responses out, embedding the NLP similarity match directly into context."""
126
- if not history:
127
- yield history
128
  return
129
 
130
- user_query = extract_text(history[-1]["content"])
131
- matched_faq, match_score = perform_nlp_cosine_matching(user_query)
132
-
133
- faq_context_injection = ""
134
-
135
- # RECTIFIED LOGIC: The metadata line only outlines telemetry metrics. It doesn't dump the duplicate answer text block.
136
- if matched_faq and match_score > 0.20:
137
- nlp_metadata_prefix = f"🔍 *NLP Intent Match: \"{matched_faq['q']}\" (Confidence: {match_score*100:.1f}%)*\n\n"
138
- faq_context_injection = (
139
- f"The user's query closely matches this curated FAQ entry:\n"
140
- f"Question: {matched_faq['q']}\n"
141
- f"Answer: {matched_faq['a']}\n"
142
- f"Please output this answer clearly and conversationally. Do not wrap it in quotes."
143
- )
144
- else:
145
- nlp_metadata_prefix = "🔍 *NLP Search Notice: No close database FAQ match found. Processing via general knowledge base...*\n\n"
146
- faq_context_injection = "No matching database entry found. Answer the question directly using standard general system knowledge."
147
 
148
- full_system_prompt = (
149
- f"{system_message}\n\n"
150
- f"{faq_context_injection}"
 
 
 
 
 
 
 
 
 
151
  )
152
-
153
- messages = [{"role": "system", "content": full_system_prompt}]
154
- clean_history = []
155
-
156
- for msg in history[:-1]:
157
- if isinstance(msg, dict):
158
- role, content = msg.get("role"), extract_text(msg.get("content"))
159
- else:
160
- role, content = getattr(msg, "role", None), extract_text(getattr(msg, "content", ""))
161
- if role and content:
162
- messages.append({"role": role, "content": content})
163
- clean_history.append({"role": role, "content": content})
164
 
165
- messages.append({"role": "user", "content": user_query})
166
- clean_history.append({"role": "user", "content": user_query})
167
 
168
- # Pre-populate the visualization header block securely
169
- clean_history.append({"role": "assistant", "content": nlp_metadata_prefix})
 
 
170
 
 
 
 
 
 
 
171
  try:
172
- for message_chunk in client.chat_completion(
173
- messages,
174
- max_tokens=int(max_tokens),
175
- stream=True,
176
- temperature=float(temperature),
177
- top_p=float(top_p),
178
- ):
179
- token = message_chunk.choices[0].delta.content
180
- if token is not None:
181
- clean_history[-1]["content"] += token
182
- yield clean_history
183
  except Exception as e:
184
- clean_history[-1]["content"] += f"\nAn error occurred while connecting to the model API: {str(e)}"
185
- yield clean_history
186
 
187
- # Custom UI Theme Stylesheet
188
  custom_css = """
189
- #header-container {
190
- background-color: #eef2f5 !important;
191
- border-radius: 12px !important;
192
- padding: 18px !important;
193
- margin-bottom: 20px !important;
194
- border: 1px solid #dcdfe6 !important;
195
- }
196
- body.dark #header-container {
197
- background-color: #1f2937 !important;
198
- border-color: #374151 !important;
199
- }
200
- #header-container * { color: #1a1a1a !important; }
201
- body.dark #header-container * { color: #f9fafb !important; }
202
- #right-meta-card {
203
- background-color: #005f53 !important;
204
- border-radius: 12px !important;
205
- padding: 24px !important;
206
- color: white !important;
207
- text-align: center;
208
- }
209
- #right-meta-card * { color: white !important; }
210
- #send-btn {
211
- background-color: #ff6600 !important;
212
- color: white !important;
213
- border: none !important;
214
- font-weight: bold !important;
215
- }
216
- #toggle-mode-btn {
217
- background-color: #4b5563 !important;
218
- color: white !important;
219
- border-radius: 20px !important;
220
- cursor: pointer;
221
- font-size: 13px !important;
222
- }
223
- .meta-status-text {
224
- text-align: left;
225
- font-size: 13px;
226
- line-height: 1.8;
227
- margin-top: 20px;
228
- border-top: 1px solid rgba(255,255,255,0.2);
229
- padding-top: 15px;
230
- }
231
  """
232
 
233
- with gr.Blocks(css=custom_css) as demo:
 
 
 
 
 
 
 
234
 
235
- # 1. Header Section
236
- with gr.Row(elem_id="header-container"):
237
- with gr.Column(scale=3):
238
- gr.Markdown(
239
- """
240
- ### About Us
241
- At FAQ Agentic Hub, we are passionate about the power of structured, NLP-driven automation and its ability to transform customer workflows. This engine utilizes real-time Tokenization and Cosine Similarity computations.
242
- """
243
- )
244
- with gr.Column(scale=1):
245
- time_display = gr.HTML(value=get_thematic_datetime())
246
- toggle_btn = gr.Button("☀️/🌙 ➔ Change Mode (Day / Night)", elem_id="toggle-mode-btn")
247
- toggle_btn.click(None, None, None, js="() => document.body.classList.toggle('dark')")
248
 
249
- # 2. Workspace Layout
250
  with gr.Row():
251
- with gr.Column(scale=2):
252
- chatbot = gr.Chatbot(
253
- label="Lane NLP AI Support Stream",
254
- value=[{"role": "assistant", "content": "Hello! I'm Lane, your upgraded NLP FAQ Support Agent. Ask me questions about shipping, returns, or certificates, and watch the cosine-similarity system extract matches in real-time."}]
 
255
  )
256
 
257
  with gr.Row():
258
- msg_input = gr.Textbox(
259
- placeholder="Ask Lane anything regarding automated FAQ policies...",
260
- show_label=False,
261
- scale=4
262
- )
263
- send_btn = gr.Button("Send", variant="primary", scale=1, elem_id="send-btn")
264
-
265
- with gr.Row():
266
- suggest_btn1 = gr.Button("What are your shipping times and policies?")
267
- suggest_btn2 = gr.Button("How do I unlock my course completion certificate?")
268
-
269
- with gr.Column(scale=1):
270
- with gr.Group(elem_id="right-meta-card"):
271
- gr.HTML(
272
- """
273
- <div style='display: flex; justify-content: center; margin-bottom: 12px;'>
274
- <div style='background: white; border-radius: 50%; padding: 12px; width: 110px; height: 110px; display: flex; align-items: center; justify-content: center; box-shadow: 0 4px 6px rgba(0,0,0,0.1);'>
275
- <img src='https://fonts.gstatic.com/s/i/short-term/release/googlesymbols/robot/default/48px.svg' style='width: 70px; filter: invert(27%) sepia(91%) saturate(2853%) hue-rotate(152deg) brightness(93%) contrast(101%);'>
276
- </div>
277
- </div>
278
- <h2 style='margin: 5px 0 2px 0; font-size: 24px;'>Lane <span style='font-size: 11px; background: rgba(255,255,255,0.25); padding: 2px 7px; border-radius: 4px; vertical-align: middle;'>NLP AI</span></h2>
279
- <p style='margin: 0; opacity: 0.85; font-size: 14px;'>Similarity-Augmented Assistant</p>
280
-
281
- <div class='meta-status-text'>
282
- <p>🟢 NLP Match Engine Online</p>
283
- <p><strong>Vectorization Model:</strong> Text TF-IDF Vectorizer</p>
284
- <p><strong>Processing Hardware:</strong> Base CPU Node</p>
285
- </div>
286
- """
287
- )
288
 
289
- with gr.Accordion("Advanced Agent Parameter Configurations", open=False):
290
- system_msg = gr.Textbox(
291
- value="You are an expert AI FAQ Agent assistant. Answer user queries concisely, expanding intelligently upon matched context summaries when available.",
292
- label="System Prompt Context Injection"
293
- )
294
- max_tokens = gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens")
295
- temperature = gr.Slider(minimum=0.1, maximum=4.0, value=0.4, step=0.1, label="Temperature")
296
- top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (Nucleus Sampling)")
 
 
 
 
297
 
298
- # 3. Components Event Wiring
299
  send_btn.click(
300
- user_side, inputs=[msg_input, chatbot], outputs=[msg_input, chatbot], queue=False
301
- ).then(
302
- bot_side, inputs=[chatbot, system_msg, max_tokens, temperature, top_p], outputs=[chatbot]
303
- )
304
-
305
- msg_input.submit(
306
- user_side, inputs=[msg_input, chatbot], outputs=[msg_input, chatbot], queue=False
307
- ).then(
308
- bot_side, inputs=[chatbot, system_msg, max_tokens, temperature, top_p], outputs=[chatbot]
309
- )
310
 
311
- suggest_btn1.click(
312
- user_side, inputs=[suggest_btn1, chatbot], outputs=[msg_input, chatbot], queue=False
313
- ).then(
314
- bot_side, inputs=[chatbot, system_msg, max_tokens, temperature, top_p], outputs=[chatbot]
315
- )
316
-
317
- suggest_btn2.click(
318
- user_side, inputs=[suggest_btn2, chatbot], outputs=[msg_input, chatbot], queue=False
319
- ).then(
320
- bot_side, inputs=[chatbot, system_msg, max_tokens, temperature, top_p], outputs=[chatbot]
321
- )
322
 
323
- # Running digital time clock update loop
324
- clock_timer = gr.Timer(1.0)
325
- clock_timer.tick(fn=get_thematic_datetime, outputs=time_display, queue=False)
326
 
327
  if __name__ == "__main__":
328
- demo.launch(theme=gr.themes.Default())
 
1
+ import os
2
  import gradio as gr
3
  from huggingface_hub import InferenceClient
4
+ from database import db
5
+
6
+ # Initialize HF Serverless Inference Client (Uses HF API, 0 CPU overhead)
7
+ HF_TOKEN = os.getenv("HF_TOKEN", None)
8
+ client = InferenceClient("Qwen/Qwen2.5-7B-Instruct", token=HF_TOKEN)
9
+
10
+ def format_graph_context(records):
11
+ """Formats retrieved graph records into readable LLM context."""
12
+ if not records:
13
+ return "No matching nodes found in Knowledge Graph.", ""
14
+
15
+ context_str = "RELEVANT KNOWLEDGE GRAPH NEIGHBORHOOD:\n"
16
+ formatted_html = "<div style='font-size: 13px; line-height: 1.6;'>"
17
+
18
+ for r in records:
19
+ context_str += f"- Article: {r['title']}\n"
20
+ context_str += f" Content: {r['content']}\n"
21
+ if r.get('category'):
22
+ context_str += f" Category: {r['category']}\n"
23
+ if r.get('errorCode'):
24
+ context_str += f" Resolves Error: {r['errorCode']} ({r.get('issueDetail', '')})\n"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
25
 
26
+ # Multi-hop prerequisites
27
+ prereqs = [p['title'] for p in r.get('prerequisites', []) if p.get('title')]
28
+ if prereqs:
29
+ context_str += f" Prerequisite Action Required: {', '.join(prereqs)}\n"
30
+ context_str += "\n"
31
+
32
+ # UI Visual Debugger Content
33
+ formatted_html += f"<b>📄 Article:</b> {r['title']}<br/>"
34
+ formatted_html += f"<b>📂 Category:</b> {r.get('category', 'N/A')}<br/>"
35
+ if prereqs:
36
+ formatted_html += f"<b>🔗 2-Hop Prerequisite:</b> <span style='color: #d97706;'>{', '.join(prereqs)}</span><br/>"
37
+ formatted_html += "<hr style='margin: 8px 0; border-color: #eee;'/>"
38
+
39
+ formatted_html += "</div>"
40
+ return context_str, formatted_html
41
+
42
+ def bot_respond(user_message, history):
43
+ """Processes user input, queries CognoDB, and streams LLM response."""
 
 
 
 
44
  if not user_message.strip():
45
+ yield history, "<i>No query entered.</i>"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
46
  return
47
 
48
+ # Step 1: Multi-Hop Cypher Traversal on CognoDB
49
+ records, db_error = db.query_graph_neighborhood(user_message)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
50
 
51
+ if db_error:
52
+ graph_context = f"Database Alert: {db_error}"
53
+ ui_debug = f"<span style='color:red;'>⚠️ {db_error}</span>"
54
+ else:
55
+ graph_context, ui_debug = format_graph_context(records)
56
+
57
+ # Step 2: Construct Prompt with Graph Context Injection
58
+ system_prompt = (
59
+ "You are the Wexa Knowledge Graph Assistant. "
60
+ "Answer the user's question accurately and concisely using ONLY the retrieved Graph Context below. "
61
+ "If prerequisites are mentioned in the context, explicitly state them to the user.\n\n"
62
+ f"{graph_context}"
63
  )
 
 
 
 
 
 
 
 
 
 
 
 
64
 
65
+ messages = [{"role": "system", "content": system_prompt}]
 
66
 
67
+ # Reconstruct conversation history
68
+ for msg in history:
69
+ messages.append(msg)
70
+ messages.append({"role": "user", "content": user_message})
71
 
72
+ # Update chatbot UI state
73
+ history.append({"role": "user", "content": user_message})
74
+ history.append({"role": "assistant", "content": ""})
75
+ yield history, ui_debug
76
+
77
+ # Step 3: Stream Response from Hugging Face Inference API
78
  try:
79
+ accumulated_text = ""
80
+ for chunk in client.chat_completion(messages, max_tokens=512, stream=True, temperature=0.3):
81
+ token = chunk.choices[0].delta.content
82
+ if token:
83
+ accumulated_text += token
84
+ history[-1]["content"] = accumulated_text
85
+ yield history, ui_debug
 
 
 
 
86
  except Exception as e:
87
+ history[-1]["content"] = f"An error occurred while generating the response: {str(e)}"
88
+ yield history, ui_debug
89
 
90
+ # --- Gradio UI Layout ---
91
  custom_css = """
92
+ #header { text-align: center; margin-bottom: 10px; }
93
+ #status-badge { padding: 8px 12px; border-radius: 6px; font-weight: bold; margin-bottom: 10px; }
94
+ .online { background-color: #dcfce7; color: #15803d; border: 1px solid #86efac; }
95
+ .offline { background-color: #fee2e2; color: #b91c1c; border: 1px solid #fca5a5; }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
96
  """
97
 
98
+ with gr.Blocks(theme=gr.themes.Soft(), css=custom_css) as demo:
99
+ gr.Markdown("# 🕸️ GraphRAG FAQ AI Agent", elem_id="header")
100
+ gr.Markdown("An intelligent support assistant powered by **CognoDB Cloud (Graph Database)** and **Qwen2.5-7B**.")
101
+
102
+ # Live Database Status Indicator
103
+ is_online = db.is_connected()
104
+ status_class = "online" if is_online else "offline"
105
+ status_text = "🟢 CognoDB Cloud: Connected & Ready" if is_online else "🔴 CognoDB Cloud: Unreachable (Fallback Mode)"
106
 
107
+ gr.HTML(f"<div id='status-badge' class='{status_class}'>{status_text}</div>")
 
 
 
 
 
 
 
 
 
 
 
 
108
 
 
109
  with gr.Row():
110
+ with gr.Column(scale=3):
111
+ chatbot = gr.Chatbot(type="messages", height=420)
112
+ msg_input = gr.Textbox(
113
+ placeholder="Ask a question (e.g., 'How do I fix 401 unauthorized errors?')...",
114
+ show_label=False
115
  )
116
 
117
  with gr.Row():
118
+ send_btn = gr.Button("Submit Query", variant="primary")
119
+ clear_btn = gr.Button("Clear Chat")
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
120
 
121
+ gr.Examples(
122
+ examples=[
123
+ "How do I fix 401 unauthorized errors?",
124
+ "How can I rotate my API keys?",
125
+ "How does graph context optimize token economics?"
126
+ ],
127
+ inputs=msg_input
128
+ )
129
+
130
+ with gr.Column(scale=2):
131
+ gr.Markdown("### 🔍 Live Graph Retrieval Inspector")
132
+ graph_inspector = gr.HTML("<i>Submit a query to inspect live Cypher 2-hop traversal context...</i>")
133
 
134
+ # Event Wiring
135
  send_btn.click(
136
+ bot_respond,
137
+ inputs=[msg_input, chatbot],
138
+ outputs=[chatbot, graph_inspector]
139
+ ).then(lambda: "", None, msg_input)
 
 
 
 
 
 
140
 
141
+ msg_input.submit(
142
+ bot_respond,
143
+ inputs=[msg_input, chatbot],
144
+ outputs=[chatbot, graph_inspector]
145
+ ).then(lambda: "", None, msg_input)
 
 
 
 
 
 
146
 
147
+ clear_btn.click(lambda: ([], "<i>Chat cleared.</i>"), None, [chatbot, graph_inspector])
 
 
148
 
149
  if __name__ == "__main__":
150
+ demo.launch()