MBG0903 commited on
Commit
6ab90a8
·
verified ·
1 Parent(s): 2b6dbcc

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +271 -396
app.py CHANGED
@@ -1,6 +1,7 @@
1
  # ================================================================
2
  # Wine Quality Capstone - Interactive RAG Chatbot
3
- # Hugging Face Spaces + Gradio
 
4
  # ================================================================
5
 
6
  import os
@@ -20,53 +21,71 @@ from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
20
 
21
  APP_TITLE = "🍷 Wine Quality Research Assistant"
22
 
23
- # IMPORTANT:
24
- # Upload this PDF into the SAME root folder as app.py
25
- PDF_FILE = "Mahesh_Govindaraj_QM640_Data_Analytics_Capstone_Project_Synopsis.pdf"
26
-
27
  EMBEDDING_MODEL_ID = "sentence-transformers/all-MiniLM-L6-v2"
28
 
29
- # Lightweight model suitable for Hugging Face CPU Space
30
  LLM_MODEL_ID = "google/flan-t5-small"
31
 
32
- TOP_K = 4
33
- CHUNK_SIZE = 800
34
- CHUNK_OVERLAP = 120
35
 
36
  DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
37
 
38
 
 
 
 
 
 
 
 
 
39
  # ================================================================
40
- # 2. VERIFY PDF
41
  # ================================================================
42
 
43
- print("=" * 60)
44
- print("Starting Wine Quality Research Assistant")
45
- print("=" * 60)
 
46
 
47
- print(f"Running on device: {DEVICE}")
48
- print(f"Looking for PDF: {PDF_FILE}")
49
 
 
 
 
 
 
50
 
51
- if not os.path.exists(PDF_FILE):
52
 
53
- print("\nFiles currently available in Space:")
54
 
55
- for file in os.listdir("."):
56
- print(" -", file)
57
 
58
- raise FileNotFoundError(
59
- f"\nPDF file not found: {PDF_FILE}\n"
60
- "Please upload the PDF to the root folder of your "
61
- "Hugging Face Space."
62
- )
 
 
63
 
 
 
64
 
65
- print(f"✅ PDF found successfully: {PDF_FILE}")
 
 
 
 
 
66
 
67
 
68
  # ================================================================
69
- # 3. READ PDF
70
  # ================================================================
71
 
72
  def clean_text(text):
@@ -76,15 +95,19 @@ def clean_text(text):
76
 
77
  text = text.replace("\x00", " ")
78
 
79
- # Remove excessive whitespace
80
  text = re.sub(r"\s+", " ", text)
81
 
82
  return text.strip()
83
 
84
 
 
 
 
 
85
  def read_pdf(pdf_path):
86
 
87
- print("\nReading PDF...")
88
 
89
  reader = PdfReader(pdf_path)
90
 
@@ -92,17 +115,26 @@ def read_pdf(pdf_path):
92
 
93
  for page_number, page in enumerate(reader.pages, start=1):
94
 
95
- text = page.extract_text()
96
 
97
- text = clean_text(text)
98
 
99
- if text:
100
 
101
- pages.append(
102
- {
103
- "page": page_number,
104
- "text": text
105
- }
 
 
 
 
 
 
 
 
 
106
  )
107
 
108
  return pages
@@ -110,18 +142,22 @@ def read_pdf(pdf_path):
110
 
111
  pages = read_pdf(PDF_FILE)
112
 
113
- print(f"✅ Successfully extracted text from {len(pages)} pages.")
114
-
115
 
116
- if len(pages) == 0:
117
 
118
  raise ValueError(
119
- "No readable text could be extracted from the PDF."
 
120
  )
121
 
122
 
 
 
 
 
 
123
  # ================================================================
124
- # 4. CREATE DOCUMENT CHUNKS
125
  # ================================================================
126
 
127
  def create_chunks(pages):
@@ -130,9 +166,10 @@ def create_chunks(pages):
130
 
131
  for page_data in pages:
132
 
133
- page_number = page_data["page"]
134
  text = page_data["text"]
135
 
 
 
136
  start = 0
137
 
138
  while start < len(text):
@@ -157,14 +194,17 @@ def create_chunks(pages):
157
 
158
  chunks = create_chunks(pages)
159
 
160
- print(f"✅ Created {len(chunks)} searchable knowledge chunks.")
 
 
 
161
 
162
 
163
  # ================================================================
164
- # 5. LOAD EMBEDDING MODEL
165
  # ================================================================
166
 
167
- print("\nLoading embedding model...")
168
 
169
  embedding_model = SentenceTransformer(
170
  EMBEDDING_MODEL_ID,
@@ -175,45 +215,40 @@ print("✅ Embedding model loaded.")
175
 
176
 
177
  # ================================================================
178
- # 6. CREATE DOCUMENT EMBEDDINGS
179
  # ================================================================
180
 
181
- print("\nCreating document embeddings...")
182
 
183
 
184
  document_texts = [
185
-
186
  chunk["text"]
187
-
188
  for chunk in chunks
189
-
190
  ]
191
 
192
 
193
  document_embeddings = embedding_model.encode(
194
-
195
  document_texts,
196
-
197
  convert_to_numpy=True,
198
-
199
  normalize_embeddings=True,
200
-
201
  show_progress_bar=False
202
-
203
  )
204
 
205
 
206
  print(
207
- f"✅ Created embeddings for "
208
- f"{len(document_embeddings)} document chunks."
209
  )
210
 
211
 
212
  # ================================================================
213
- # 7. LOAD FLAN-T5
214
  # ================================================================
215
 
216
- print(f"\nLoading language model: {LLM_MODEL_ID}")
 
 
 
217
 
218
 
219
  tokenizer = AutoTokenizer.from_pretrained(
@@ -231,32 +266,25 @@ llm_model.to(DEVICE)
231
  llm_model.eval()
232
 
233
 
234
- print("✅ Language model loaded.")
235
 
236
 
237
  # ================================================================
238
- # 8. SEMANTIC SEARCH
239
  # ================================================================
240
 
241
  def retrieve_context(question, top_k=TOP_K):
242
 
243
  question_embedding = embedding_model.encode(
244
-
245
  [question],
246
-
247
  convert_to_numpy=True,
248
-
249
  normalize_embeddings=True
250
-
251
  )[0]
252
 
253
 
254
  similarity_scores = np.dot(
255
-
256
  document_embeddings,
257
-
258
  question_embedding
259
-
260
  )
261
 
262
 
@@ -273,9 +301,7 @@ def retrieve_context(question, top_k=TOP_K):
273
  results.append(
274
  {
275
  "page": chunks[index]["page"],
276
-
277
  "text": chunks[index]["text"],
278
-
279
  "score": float(
280
  similarity_scores[index]
281
  )
@@ -287,31 +313,28 @@ def retrieve_context(question, top_k=TOP_K):
287
 
288
 
289
  # ================================================================
290
- # 9. BUILD RETRIEVED CONTEXT
291
  # ================================================================
292
 
293
  def build_context(retrieved_chunks):
294
 
295
  context_parts = []
296
 
297
-
298
  for item in retrieved_chunks:
299
 
300
  context_parts.append(
301
-
302
  f"""
303
- SOURCE PAGE {item['page']}
304
 
305
  {item['text']}
306
  """
307
  )
308
 
309
-
310
- return "\n\n".join(context_parts)
311
 
312
 
313
  # ================================================================
314
- # 10. FORMAT CHAT HISTORY
315
  # ================================================================
316
 
317
  def format_history(history):
@@ -321,126 +344,116 @@ def format_history(history):
321
  return "No previous conversation."
322
 
323
 
324
- conversation = []
325
 
 
 
326
 
327
- # Only use recent messages
328
- recent_history = history[-4:]
329
 
330
-
331
- for item in recent_history:
332
 
333
  if isinstance(item, dict):
334
 
335
- role = item.get("role", "")
 
 
 
336
 
337
- content = item.get("content", "")
 
 
 
338
 
339
- conversation.append(
340
  f"{role}: {content}"
341
  )
342
 
343
 
344
- return "\n".join(conversation)
345
 
346
 
347
  # ================================================================
348
- # 11. GENERATE ANSWER
349
  # ================================================================
350
 
351
- def generate_answer(question, context, history_text):
 
 
 
 
352
 
353
  prompt = f"""
354
- You are an intelligent academic research assistant for the
355
- Data Analytics Capstone project:
356
 
357
- Machine Learning-Based Prediction of Wine Quality Using
358
- Physicochemical Properties.
359
 
360
- The research document belongs to Mahesh Govindaraj.
361
 
362
- Your job is to answer questions based ONLY on the supplied
363
- research document.
364
 
365
- IMPORTANT RULES:
366
 
367
- 1. Use the DOCUMENT CONTEXT as the primary source.
368
- 2. Do not invent information.
369
- 3. Do not invent model results.
370
- 4. Clearly distinguish planned methodology from completed work.
371
- 5. Explain technical concepts in simple language when appropriate.
372
- 6. Give concise but informative answers.
373
- 7. Use bullet points when they improve readability.
374
- 8. If the answer is not available in the document, say:
375
-
376
- "This information is not specified in the current capstone synopsis."
377
-
378
- 9. Do not claim that a model performed best unless the document
379
- explicitly provides that result.
380
- 10. When discussing the dataset, models, research questions,
381
- methodology or business value, remain consistent with the
382
- supplied document.
383
 
 
384
 
385
  DOCUMENT CONTEXT:
386
 
387
  {context}
388
 
389
-
390
  RECENT CONVERSATION:
391
 
392
- {history_text}
393
-
394
 
395
  USER QUESTION:
396
 
397
  {question}
398
 
399
-
400
  ANSWER:
401
  """
402
 
403
 
404
  inputs = tokenizer(
405
-
406
  prompt,
407
-
408
  return_tensors="pt",
409
-
410
  truncation=True,
 
 
411
 
412
- max_length=2048
413
 
414
- ).to(DEVICE)
 
 
 
415
 
416
 
417
  with torch.no_grad():
418
 
419
  outputs = llm_model.generate(
420
-
421
  **inputs,
422
-
423
- max_new_tokens=220,
424
-
425
  do_sample=False,
426
-
427
- num_beams=3,
428
-
429
  repetition_penalty=1.15,
430
-
431
  no_repeat_ngram_size=3,
432
-
433
  early_stopping=True
434
-
435
  )
436
 
437
 
438
  answer = tokenizer.decode(
439
-
440
  outputs[0],
441
-
442
  skip_special_tokens=True
443
-
444
  )
445
 
446
 
@@ -448,41 +461,46 @@ ANSWER:
448
 
449
 
450
  # ================================================================
451
- # 12. MAIN CHATBOT FUNCTION
452
  # ================================================================
453
 
454
- def chatbot_response(message, history):
455
 
456
  if not message or not message.strip():
457
 
458
- return "Please enter a question about the capstone project."
 
 
 
 
459
 
 
460
 
461
- # ------------------------------------------------------------
462
- # Retrieve relevant document content
463
- # ------------------------------------------------------------
464
 
465
- retrieved_chunks = retrieve_context(message)
466
 
 
 
 
467
 
468
- # Debug information in HF logs
469
- print("\n" + "=" * 60)
 
470
 
471
- print(f"QUESTION: {message}")
472
 
473
- print("Retrieved document pages:")
474
 
475
 
476
  for item in retrieved_chunks:
477
 
478
  print(
479
- f"Page {item['page']} "
480
- f"| Similarity: {item['score']:.3f}"
481
  )
482
 
483
 
484
  # ------------------------------------------------------------
485
- # Build LLM context
486
  # ------------------------------------------------------------
487
 
488
  context = build_context(
@@ -490,7 +508,7 @@ def chatbot_response(message, history):
490
  )
491
 
492
 
493
- history_text = format_history(
494
  history
495
  )
496
 
@@ -500,138 +518,114 @@ def chatbot_response(message, history):
500
  # ------------------------------------------------------------
501
 
502
  answer = generate_answer(
503
-
504
  message,
505
-
506
  context,
507
-
508
- history_text
509
-
510
  )
511
 
512
 
513
  # ------------------------------------------------------------
514
- # Extract unique source pages
515
  # ------------------------------------------------------------
516
 
517
  source_pages = sorted(
518
-
519
  set(
520
-
521
  item["page"]
522
-
523
  for item in retrieved_chunks
524
-
525
  )
526
-
527
  )
528
 
529
 
530
- page_list = ", ".join(
531
-
532
  str(page)
533
-
534
  for page in source_pages
535
-
536
  )
537
 
538
 
539
- # ------------------------------------------------------------
540
- # Add source information
541
- # ------------------------------------------------------------
542
-
543
- final_answer = f"""
544
  {answer}
545
 
546
  ---
547
 
548
- 📚 **Source pages:** {page_list}
549
 
550
- 📄 **Knowledge source:** QM640 Data Analytics Capstone Synopsis
551
  """
552
 
553
 
554
- return final_answer
555
-
556
-
557
  # ================================================================
558
- # 13. GRADIO INTERFACE
559
  # ================================================================
560
 
561
- custom_css = """
562
 
563
  .gradio-container {
564
  max-width: 1100px !important;
565
  margin: auto !important;
566
  }
567
 
568
- footer {
569
- visibility: hidden;
570
- }
571
-
572
  """
573
 
574
 
575
  with gr.Blocks(
576
-
577
  title=APP_TITLE,
578
-
579
- css=custom_css
580
-
581
  ) as demo:
582
 
583
 
584
- # ------------------------------------------------------------
585
- # Header
586
- # ------------------------------------------------------------
587
-
588
  gr.Markdown(
589
  """
590
  # 🍷 Wine Quality Research Assistant
591
 
592
  ### Machine Learning-Based Prediction of Wine Quality Using Physicochemical Properties
593
 
594
- Ask questions about the **QM640 Data Analytics Capstone Project**.
595
-
596
- This AI assistant searches the uploaded research synopsis before
597
- generating an answer.
598
-
599
- ---
600
-
601
- ### 💡 You can ask about
602
-
603
- **Research**
604
- • Problem Statement
605
- • Objectives
606
- • Research Questions
607
- • Hypotheses
608
-
609
- **Dataset**
610
- • Wine Quality Dataset
611
- • Predictor Variables
612
- • Target Variable
613
- • Sample Size
614
-
615
- **Machine Learning**
616
- • Linear Regression
617
- • Decision Tree
618
- • Random Forest
619
- • Gradient Boosting
620
- • XGBoost
621
-
622
- **Evaluation & Explainability**
623
- • MAE
624
- • RMSE
625
- • R²
626
- • Cross Validation
627
- • Hyperparameter Optimisation
628
- • SHAP
629
- • LIME
630
-
631
- **Business**
632
- • Business Impact
633
- • Practical Applications
634
- • Study Limitations
 
 
 
 
 
 
635
 
636
  ---
637
  """
@@ -639,104 +633,52 @@ generating an answer.
639
 
640
 
641
  # ------------------------------------------------------------
642
- # PDF status
643
  # ------------------------------------------------------------
644
 
645
  gr.Markdown(
646
  f"""
647
- ### ✅ Knowledge Base Loaded
648
-
649
- 📄 **Document:** `{PDF_FILE}`
650
 
651
- 📑 **Pages extracted:** {len(pages)}
652
 
653
- 🧩 **Knowledge chunks:** {len(chunks)}
654
 
655
- 🤖 **Language model:** `{LLM_MODEL_ID}`
656
 
657
  🔎 **Embedding model:** `{EMBEDDING_MODEL_ID}`
 
 
 
 
658
  """
659
  )
660
 
661
 
662
  # ------------------------------------------------------------
663
- # Chat window
664
  # ------------------------------------------------------------
665
 
666
- chatbot_ui = gr.Chatbot(
667
-
668
  type="messages",
669
-
670
  height=500,
671
-
672
- show_label=False,
673
-
674
- placeholder="""
675
- 🍷 Welcome to the Wine Quality Research Assistant!
676
-
677
- Try asking:
678
-
679
- "What is the main objective of the study?"
680
-
681
- "What dataset is being used?"
682
-
683
- "What are the research questions?"
684
-
685
- "Which machine learning models will be compared?"
686
-
687
- "How will SHAP be used?"
688
- """
689
-
690
  )
691
 
692
 
693
- # ------------------------------------------------------------
694
- # Input area
695
- # ------------------------------------------------------------
696
-
697
- with gr.Row():
698
-
699
- user_input = gr.Textbox(
700
-
701
- placeholder=
702
- "Ask a question about the Wine Quality Capstone...",
703
-
704
- show_label=False,
705
-
706
- scale=8
707
-
708
- )
709
-
710
-
711
- submit_button = gr.Button(
712
-
713
- "Ask 🍷",
714
-
715
- variant="primary",
716
-
717
- scale=1
718
-
719
- )
720
-
721
-
722
- clear_button = gr.Button(
723
- "🗑️ Clear Conversation"
724
- )
725
-
726
 
727
- # ------------------------------------------------------------
728
- # Example questions
729
- # ------------------------------------------------------------
730
 
731
- gr.Examples(
732
 
733
  examples=[
734
 
735
- "What is the main objective of this research?",
736
 
737
- "What problem is this project trying to solve?",
738
 
739
- "What dataset is being used?",
740
 
741
  "How many observations are in the dataset?",
742
 
@@ -746,139 +688,72 @@ Try asking:
746
 
747
  "Explain the four research questions.",
748
 
749
- "Which machine learning models will be compared?",
750
 
751
  "How will the models be validated?",
752
 
753
- "What is 5-fold cross-validation?",
 
 
 
 
754
 
755
- "What are MAE, RMSE and R²?",
756
 
757
  "How will SHAP be used?",
758
 
759
- "What is the role of LIME?",
760
 
761
- "What is the business value of this research?",
762
 
763
  "What are the limitations of the study?"
764
 
765
- ],
766
-
767
- inputs=user_input
768
 
769
  )
770
 
771
 
772
- # ------------------------------------------------------------
773
- # Footer
774
- # ------------------------------------------------------------
775
-
776
  gr.Markdown(
777
  """
778
  ---
779
 
780
- ### 🔎 How this assistant works
781
-
782
- **Question → Semantic Search → Relevant PDF Content → AI → Answer**
783
 
784
- The chatbot uses Retrieval-Augmented Generation (RAG) to search
785
- the capstone synopsis before generating its response.
786
-
787
- *QM640 Data Analytics Capstone – Wine Quality Prediction*
788
- """
789
- )
790
 
 
791
 
792
- # ============================================================
793
- # 14. CHAT EVENTS
794
- # ============================================================
795
 
796
- def respond(message, chat_history):
797
 
798
- chat_history = chat_history or []
799
 
 
800
 
801
- response = chatbot_response(
802
 
803
- message,
804
 
805
- chat_history
806
-
807
- )
808
-
809
-
810
- chat_history.append(
811
- {
812
- "role": "user",
813
- "content": message
814
- }
815
- )
816
-
817
-
818
- chat_history.append(
819
- {
820
- "role": "assistant",
821
- "content": response
822
- }
823
- )
824
 
 
 
825
 
826
- return "", chat_history
827
-
828
-
829
- # Button click
830
- submit_button.click(
831
-
832
- respond,
833
-
834
- inputs=[
835
- user_input,
836
- chatbot_ui
837
- ],
838
-
839
- outputs=[
840
- user_input,
841
- chatbot_ui
842
- ]
843
-
844
- )
845
-
846
-
847
- # Enter key
848
- user_input.submit(
849
-
850
- respond,
851
-
852
- inputs=[
853
- user_input,
854
- chatbot_ui
855
- ],
856
-
857
- outputs=[
858
- user_input,
859
- chatbot_ui
860
- ]
861
-
862
- )
863
-
864
-
865
- # Clear
866
- clear_button.click(
867
-
868
- lambda: [],
869
-
870
- outputs=chatbot_ui
871
-
872
  )
873
 
874
 
875
  # ================================================================
876
- # 15. LAUNCH APPLICATION
877
  # ================================================================
878
 
879
- print("\n" + "=" * 60)
880
- print("✅ Wine Quality Research Assistant is ready!")
881
- print("=" * 60)
 
 
882
 
883
 
884
  if __name__ == "__main__":
 
1
  # ================================================================
2
  # Wine Quality Capstone - Interactive RAG Chatbot
3
+ # Hugging Face Spaces
4
+ # Compatible with Gradio 6.x
5
  # ================================================================
6
 
7
  import os
 
21
 
22
  APP_TITLE = "🍷 Wine Quality Research Assistant"
23
 
 
 
 
 
24
  EMBEDDING_MODEL_ID = "sentence-transformers/all-MiniLM-L6-v2"
25
 
26
+ # Small model chosen for Hugging Face CPU Spaces
27
  LLM_MODEL_ID = "google/flan-t5-small"
28
 
29
+ TOP_K = 3
30
+ CHUNK_SIZE = 700
31
+ CHUNK_OVERLAP = 100
32
 
33
  DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
34
 
35
 
36
+ print("=" * 70)
37
+ print("🍷 Starting Wine Quality Research Assistant")
38
+ print("=" * 70)
39
+
40
+ print(f"Device: {DEVICE}")
41
+ print(f"Gradio Version: {gr.__version__}")
42
+
43
+
44
  # ================================================================
45
+ # 2. FIND PDF AUTOMATICALLY
46
  # ================================================================
47
 
48
+ def find_pdf():
49
+ """
50
+ Automatically locate a PDF in the Space repository.
51
+ """
52
 
53
+ files = os.listdir(".")
 
54
 
55
+ pdf_files = [
56
+ file
57
+ for file in files
58
+ if file.lower().endswith(".pdf")
59
+ ]
60
 
61
+ if not pdf_files:
62
 
63
+ print("\nFiles found in repository:")
64
 
65
+ for file in files:
66
+ print(" -", file)
67
 
68
+ raise FileNotFoundError(
69
+ "\nNo PDF file found.\n"
70
+ "Please upload your capstone PDF into the root folder "
71
+ "of the Hugging Face Space."
72
+ )
73
+
74
+ print("\nPDF files detected:")
75
 
76
+ for pdf in pdf_files:
77
+ print(" -", pdf)
78
 
79
+ return pdf_files[0]
80
+
81
+
82
+ PDF_FILE = find_pdf()
83
+
84
+ print(f"\n✅ Using PDF: {PDF_FILE}")
85
 
86
 
87
  # ================================================================
88
+ # 3. CLEAN TEXT
89
  # ================================================================
90
 
91
  def clean_text(text):
 
95
 
96
  text = text.replace("\x00", " ")
97
 
98
+ # Fix excessive whitespace
99
  text = re.sub(r"\s+", " ", text)
100
 
101
  return text.strip()
102
 
103
 
104
+ # ================================================================
105
+ # 4. READ PDF
106
+ # ================================================================
107
+
108
  def read_pdf(pdf_path):
109
 
110
+ print("\n📖 Reading PDF...")
111
 
112
  reader = PdfReader(pdf_path)
113
 
 
115
 
116
  for page_number, page in enumerate(reader.pages, start=1):
117
 
118
+ try:
119
 
120
+ text = page.extract_text()
121
 
122
+ text = clean_text(text)
123
 
124
+ if text:
125
+
126
+ pages.append(
127
+ {
128
+ "page": page_number,
129
+ "text": text
130
+ }
131
+ )
132
+
133
+ except Exception as error:
134
+
135
+ print(
136
+ f"Warning: Could not read page "
137
+ f"{page_number}: {error}"
138
  )
139
 
140
  return pages
 
142
 
143
  pages = read_pdf(PDF_FILE)
144
 
 
 
145
 
146
+ if not pages:
147
 
148
  raise ValueError(
149
+ "The PDF was found but no readable text "
150
+ "could be extracted."
151
  )
152
 
153
 
154
+ print(
155
+ f"✅ Text extracted from {len(pages)} pages."
156
+ )
157
+
158
+
159
  # ================================================================
160
+ # 5. CREATE KNOWLEDGE CHUNKS
161
  # ================================================================
162
 
163
  def create_chunks(pages):
 
166
 
167
  for page_data in pages:
168
 
 
169
  text = page_data["text"]
170
 
171
+ page_number = page_data["page"]
172
+
173
  start = 0
174
 
175
  while start < len(text):
 
194
 
195
  chunks = create_chunks(pages)
196
 
197
+
198
+ print(
199
+ f"✅ Created {len(chunks)} searchable knowledge chunks."
200
+ )
201
 
202
 
203
  # ================================================================
204
+ # 6. LOAD EMBEDDING MODEL
205
  # ================================================================
206
 
207
+ print("\n🔎 Loading embedding model...")
208
 
209
  embedding_model = SentenceTransformer(
210
  EMBEDDING_MODEL_ID,
 
215
 
216
 
217
  # ================================================================
218
+ # 7. CREATE EMBEDDINGS
219
  # ================================================================
220
 
221
+ print("\n🧠 Creating document embeddings...")
222
 
223
 
224
  document_texts = [
 
225
  chunk["text"]
 
226
  for chunk in chunks
 
227
  ]
228
 
229
 
230
  document_embeddings = embedding_model.encode(
 
231
  document_texts,
 
232
  convert_to_numpy=True,
 
233
  normalize_embeddings=True,
 
234
  show_progress_bar=False
 
235
  )
236
 
237
 
238
  print(
239
+ f"✅ Embeddings created for "
240
+ f"{len(document_embeddings)} chunks."
241
  )
242
 
243
 
244
  # ================================================================
245
+ # 8. LOAD LANGUAGE MODEL
246
  # ================================================================
247
 
248
+ print(
249
+ f"\n🤖 Loading language model: "
250
+ f"{LLM_MODEL_ID}"
251
+ )
252
 
253
 
254
  tokenizer = AutoTokenizer.from_pretrained(
 
266
  llm_model.eval()
267
 
268
 
269
+ print("✅ Language model loaded successfully.")
270
 
271
 
272
  # ================================================================
273
+ # 9. SEMANTIC SEARCH
274
  # ================================================================
275
 
276
  def retrieve_context(question, top_k=TOP_K):
277
 
278
  question_embedding = embedding_model.encode(
 
279
  [question],
 
280
  convert_to_numpy=True,
 
281
  normalize_embeddings=True
 
282
  )[0]
283
 
284
 
285
  similarity_scores = np.dot(
 
286
  document_embeddings,
 
287
  question_embedding
 
288
  )
289
 
290
 
 
301
  results.append(
302
  {
303
  "page": chunks[index]["page"],
 
304
  "text": chunks[index]["text"],
 
305
  "score": float(
306
  similarity_scores[index]
307
  )
 
313
 
314
 
315
  # ================================================================
316
+ # 10. BUILD CONTEXT
317
  # ================================================================
318
 
319
  def build_context(retrieved_chunks):
320
 
321
  context_parts = []
322
 
 
323
  for item in retrieved_chunks:
324
 
325
  context_parts.append(
 
326
  f"""
327
+ SOURCE PAGE {item['page']}:
328
 
329
  {item['text']}
330
  """
331
  )
332
 
333
+ return "\n".join(context_parts)
 
334
 
335
 
336
  # ================================================================
337
+ # 11. PREPARE CONVERSATION HISTORY
338
  # ================================================================
339
 
340
  def format_history(history):
 
344
  return "No previous conversation."
345
 
346
 
347
+ history_text = []
348
 
349
+ # Keep only recent messages
350
+ recent_messages = history[-4:]
351
 
 
 
352
 
353
+ for item in recent_messages:
 
354
 
355
  if isinstance(item, dict):
356
 
357
+ role = item.get(
358
+ "role",
359
+ "user"
360
+ )
361
 
362
+ content = item.get(
363
+ "content",
364
+ ""
365
+ )
366
 
367
+ history_text.append(
368
  f"{role}: {content}"
369
  )
370
 
371
 
372
+ return "\n".join(history_text)
373
 
374
 
375
  # ================================================================
376
+ # 12. GENERATE RESPONSE
377
  # ================================================================
378
 
379
+ def generate_answer(
380
+ question,
381
+ context,
382
+ conversation_history
383
+ ):
384
 
385
  prompt = f"""
386
+ You are an academic research assistant for the capstone project:
 
387
 
388
+ "Machine Learning-Based Prediction of Wine Quality Using
389
+ Physicochemical Properties."
390
 
391
+ The project is based on the supplied capstone synopsis.
392
 
393
+ Your task is to answer the user's question using the DOCUMENT
394
+ CONTEXT below.
395
 
396
+ IMPORTANT INSTRUCTIONS:
397
 
398
+ - Answer based primarily on the supplied document.
399
+ - Do not invent facts.
400
+ - Do not invent model results.
401
+ - Do not say that a planned analysis has already been completed.
402
+ - Clearly distinguish planned methodology from completed findings.
403
+ - Explain technical concepts in simple language.
404
+ - Keep answers clear and reasonably concise.
405
+ - Use bullet points when useful.
406
+ - If the document does not contain the requested information,
407
+ clearly say:
 
 
 
 
 
 
408
 
409
+ "This information is not specified in the current capstone synopsis."
410
 
411
  DOCUMENT CONTEXT:
412
 
413
  {context}
414
 
 
415
  RECENT CONVERSATION:
416
 
417
+ {conversation_history}
 
418
 
419
  USER QUESTION:
420
 
421
  {question}
422
 
 
423
  ANSWER:
424
  """
425
 
426
 
427
  inputs = tokenizer(
 
428
  prompt,
 
429
  return_tensors="pt",
 
430
  truncation=True,
431
+ max_length=1800
432
+ )
433
 
 
434
 
435
+ inputs = {
436
+ key: value.to(DEVICE)
437
+ for key, value in inputs.items()
438
+ }
439
 
440
 
441
  with torch.no_grad():
442
 
443
  outputs = llm_model.generate(
 
444
  **inputs,
445
+ max_new_tokens=200,
 
 
446
  do_sample=False,
447
+ num_beams=2,
 
 
448
  repetition_penalty=1.15,
 
449
  no_repeat_ngram_size=3,
 
450
  early_stopping=True
 
451
  )
452
 
453
 
454
  answer = tokenizer.decode(
 
455
  outputs[0],
 
456
  skip_special_tokens=True
 
457
  )
458
 
459
 
 
461
 
462
 
463
  # ================================================================
464
+ # 13. MAIN CHAT FUNCTION
465
  # ================================================================
466
 
467
+ def chat(message, history):
468
 
469
  if not message or not message.strip():
470
 
471
+ return (
472
+ "Please enter a question about "
473
+ "the capstone project."
474
+ )
475
+
476
 
477
+ print("\n" + "=" * 70)
478
 
479
+ print(f"Question: {message}")
 
 
480
 
 
481
 
482
+ # ------------------------------------------------------------
483
+ # Retrieve relevant PDF chunks
484
+ # ------------------------------------------------------------
485
 
486
+ retrieved_chunks = retrieve_context(
487
+ message
488
+ )
489
 
 
490
 
491
+ print("\nRelevant pages retrieved:")
492
 
493
 
494
  for item in retrieved_chunks:
495
 
496
  print(
497
+ f"Page {item['page']} | "
498
+ f"Similarity: {item['score']:.3f}"
499
  )
500
 
501
 
502
  # ------------------------------------------------------------
503
+ # Create context
504
  # ------------------------------------------------------------
505
 
506
  context = build_context(
 
508
  )
509
 
510
 
511
+ conversation_history = format_history(
512
  history
513
  )
514
 
 
518
  # ------------------------------------------------------------
519
 
520
  answer = generate_answer(
 
521
  message,
 
522
  context,
523
+ conversation_history
 
 
524
  )
525
 
526
 
527
  # ------------------------------------------------------------
528
+ # Source pages
529
  # ------------------------------------------------------------
530
 
531
  source_pages = sorted(
 
532
  set(
 
533
  item["page"]
 
534
  for item in retrieved_chunks
 
535
  )
 
536
  )
537
 
538
 
539
+ pages_string = ", ".join(
 
540
  str(page)
 
541
  for page in source_pages
 
542
  )
543
 
544
 
545
+ return f"""
 
 
 
 
546
  {answer}
547
 
548
  ---
549
 
550
+ 📚 **Relevant source page(s): {pages_string}**
551
 
552
+ 📄 *Answer generated from the uploaded QM640 Capstone Synopsis.*
553
  """
554
 
555
 
 
 
 
556
  # ================================================================
557
+ # 14. GRADIO INTERFACE
558
  # ================================================================
559
 
560
+ css = """
561
 
562
  .gradio-container {
563
  max-width: 1100px !important;
564
  margin: auto !important;
565
  }
566
 
 
 
 
 
567
  """
568
 
569
 
570
  with gr.Blocks(
 
571
  title=APP_TITLE,
572
+ css=css
 
 
573
  ) as demo:
574
 
575
 
 
 
 
 
576
  gr.Markdown(
577
  """
578
  # 🍷 Wine Quality Research Assistant
579
 
580
  ### Machine Learning-Based Prediction of Wine Quality Using Physicochemical Properties
581
 
582
+ Welcome to the **QM640 Data Analytics Capstone Research Assistant**.
583
+
584
+ This chatbot searches the uploaded capstone synopsis and uses
585
+ the most relevant sections of the document to answer your
586
+ questions.
587
+
588
+ You can ask questions about:
589
+
590
+ **📊 Research**
591
+ - Problem Statement
592
+ - Purpose of the Study
593
+ - Research Objectives
594
+ - Research Questions
595
+ - Research Hypotheses
596
+
597
+ **🍷 Dataset**
598
+ - Wine Quality Dataset
599
+ - Physicochemical Variables
600
+ - Target Variable
601
+ - Sample Size
602
+ - Data Source
603
+
604
+ **🤖 Machine Learning**
605
+ - Linear Regression
606
+ - Decision Tree
607
+ - Random Forest
608
+ - Gradient Boosting
609
+ - XGBoost
610
+
611
+ **📈 Model Evaluation**
612
+ - MAE
613
+ - RMSE
614
+ - R²
615
+ - 80/20 Train-Test Split
616
+ - 5-Fold Cross Validation
617
+ - Hyperparameter Optimisation
618
+
619
+ **🔍 Explainable AI**
620
+ - Feature Importance
621
+ - SHAP
622
+ - LIME
623
+
624
+ **💼 Business**
625
+ - Business Impact
626
+ - Quality Assessment
627
+ - Production Optimisation
628
+ - Study Limitations
629
 
630
  ---
631
  """
 
633
 
634
 
635
  # ------------------------------------------------------------
636
+ # Show Knowledge Base Status
637
  # ------------------------------------------------------------
638
 
639
  gr.Markdown(
640
  f"""
641
+ ### ✅ Knowledge Base Status
 
 
642
 
643
+ 📄 **PDF:** `{PDF_FILE}`
644
 
645
+ 📑 **Pages loaded:** {len(pages)}
646
 
647
+ 🧩 **Searchable chunks:** {len(chunks)}
648
 
649
  🔎 **Embedding model:** `{EMBEDDING_MODEL_ID}`
650
+
651
+ 🤖 **Answer model:** `{LLM_MODEL_ID}`
652
+
653
+ 💻 **Device:** `{DEVICE}`
654
  """
655
  )
656
 
657
 
658
  # ------------------------------------------------------------
659
+ # Chat Interface
660
  # ------------------------------------------------------------
661
 
662
+ chatbot_component = gr.Chatbot(
 
663
  type="messages",
 
664
  height=500,
665
+ show_label=False
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
666
  )
667
 
668
 
669
+ chat_interface = gr.ChatInterface(
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
670
 
671
+ fn=chat,
 
 
672
 
673
+ chatbot=chatbot_component,
674
 
675
  examples=[
676
 
677
+ "What is the main objective of this study?",
678
 
679
+ "What problem is the research trying to solve?",
680
 
681
+ "What dataset is used in this project?",
682
 
683
  "How many observations are in the dataset?",
684
 
 
688
 
689
  "Explain the four research questions.",
690
 
691
+ "Which machine learning models are being compared?",
692
 
693
  "How will the models be validated?",
694
 
695
+ "Explain 5-fold cross-validation in simple terms.",
696
+
697
+ "What is MAE?",
698
+
699
+ "What is RMSE?",
700
 
701
+ "What does R² mean?",
702
 
703
  "How will SHAP be used?",
704
 
705
+ "What is LIME?",
706
 
707
+ "What is the business value of the project?",
708
 
709
  "What are the limitations of the study?"
710
 
711
+ ]
 
 
712
 
713
  )
714
 
715
 
 
 
 
 
716
  gr.Markdown(
717
  """
718
  ---
719
 
720
+ ### 🔎 How the chatbot works
 
 
721
 
722
+ **Your Question**
 
 
 
 
 
723
 
724
+ ↓
725
 
726
+ **Semantic Search**
 
 
727
 
728
+ ↓
729
 
730
+ **Find Relevant Sections of the PDF**
731
 
732
+ ↓
733
 
734
+ **Provide Those Sections to the Language Model**
735
 
736
+ ↓
737
 
738
+ **Generate the Answer**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
739
 
740
+ This is a simple **Retrieval-Augmented Generation (RAG)**
741
+ implementation.
742
 
743
+ *QM640 Data Analytics Capstone*
744
+ """
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
745
  )
746
 
747
 
748
  # ================================================================
749
+ # 15. START APPLICATION
750
  # ================================================================
751
 
752
+ print("\n" + "=" * 70)
753
+
754
+ print("✅ Wine Quality Research Assistant is ready.")
755
+
756
+ print("=" * 70)
757
 
758
 
759
  if __name__ == "__main__":