File size: 54,121 Bytes
cf02b2b
bef54cd
cf02b2b
 
 
 
 
 
 
 
 
 
 
f139d4e
 
cf02b2b
 
 
 
 
 
f139d4e
5065491
cf02b2b
 
 
 
 
 
dad5387
4a1bc78
dad5387
 
4a1bc78
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dad5387
 
 
 
4a1bc78
 
 
dad5387
 
 
 
 
 
 
 
 
4a1bc78
 
 
 
 
 
 
 
 
 
 
 
 
dad5387
 
 
ce2d17d
4a1bc78
dad5387
4a1bc78
ce2d17d
 
dad5387
 
 
ce2d17d
 
 
dad5387
 
 
 
 
4a1bc78
 
 
dad5387
 
 
 
 
 
 
 
4a1bc78
 
dad5387
 
 
 
4a1bc78
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dad5387
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9341027
dad5387
 
9341027
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dad5387
 
 
 
9341027
 
 
 
 
 
 
 
dad5387
 
9341027
dad5387
 
 
 
9341027
dad5387
 
 
 
 
 
 
9341027
 
 
 
 
 
 
 
 
 
dad5387
 
 
 
9341027
 
dad5387
 
9341027
dad5387
 
9341027
dad5387
9341027
dad5387
 
9341027
dad5387
 
 
 
9341027
 
 
dad5387
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2b377e5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cf02b2b
 
 
 
 
 
 
 
 
1eb2c60
cf02b2b
1eb2c60
cf02b2b
 
 
 
 
 
 
 
102aa44
 
 
 
 
 
 
cf02b2b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
102aa44
 
 
 
 
 
 
 
 
 
 
 
 
cf02b2b
 
 
 
 
102aa44
cf02b2b
 
 
 
 
102aa44
cf02b2b
 
102aa44
cf02b2b
 
 
 
 
102aa44
cf02b2b
 
 
 
102aa44
 
 
 
 
 
 
 
 
 
 
 
 
 
cf02b2b
 
 
 
102aa44
 
 
 
 
 
 
 
 
cf02b2b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
08a93b9
cf02b2b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
899843c
a0b0f78
36ccb7e
a0b0f78
5065491
899843c
a0b0f78
36ccb7e
899843c
 
36ccb7e
 
 
 
 
 
 
 
f139d4e
 
899843c
36ccb7e
 
 
 
 
 
cf02b2b
 
 
 
 
 
 
 
c3ade54
 
 
 
 
 
 
cf02b2b
 
 
 
 
 
 
 
 
8a78e3e
cf02b2b
 
 
 
 
 
 
 
 
8a78e3e
 
 
 
 
 
 
 
 
 
 
 
dad5387
8a78e3e
dad5387
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cf02b2b
 
 
 
 
 
75546b0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9341027
75546b0
dad5387
75546b0
9341027
 
 
 
 
 
 
 
 
 
 
 
 
75546b0
 
cf02b2b
75546b0
 
 
 
 
 
cf02b2b
2b377e5
dad5387
2b377e5
 
 
 
 
 
dad5387
2b377e5
 
cf02b2b
dad5387
cf02b2b
 
2b377e5
 
8a78e3e
2b377e5
 
 
 
cf02b2b
 
9341027
2b377e5
dad5387
9341027
2b377e5
 
9341027
2b377e5
 
 
9341027
dad5387
 
2b377e5
9341027
2b377e5
9341027
 
2b377e5
 
9341027
8a78e3e
dad5387
 
2b377e5
8a78e3e
 
cf02b2b
a0b0f78
 
5065491
899843c
a0b0f78
15b123d
899843c
 
15b123d
 
 
 
 
 
 
 
 
 
899843c
15b123d
 
 
 
 
 
 
 
 
 
a0b0f78
 
cf02b2b
8a78e3e
cf02b2b
 
 
 
 
 
 
 
 
8a78e3e
dad5387
8a78e3e
cf02b2b
 
8a78e3e
cf02b2b
 
 
 
 
8a78e3e
cf02b2b
 
 
5060335
8a78e3e
 
cf02b2b
 
 
 
 
8a78e3e
cf02b2b
 
5065491
 
cf02b2b
dad5387
 
 
 
5065491
 
 
 
 
 
 
 
 
 
 
b297e4e
5065491
82583bd
dad5387
5065491
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dad5387
 
 
ce2d17d
 
 
 
 
 
 
 
 
 
 
dad5387
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f9f45e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
061a93c
 
 
 
 
 
 
 
 
 
 
f139d4e
 
061a93c
 
bef54cd
dad5387
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bef54cd
dad5387
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cf02b2b
 
 
 
 
943f386
 
cf02b2b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
943f386
 
cf02b2b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
"""
Enhanced WebSocket handler with hybrid LLM and voice features
"""

from fastapi import WebSocket, WebSocketDisconnect
from langchain_core.messages import HumanMessage, SystemMessage, AIMessage
import logging
import json
import asyncio
import uuid
import tempfile
import base64
from pathlib import Path
import io
import matplotlib.pyplot as plt

from llm_service import create_graph, create_basic_graph
from lancedb_service import lancedb_service
from hybrid_llm_service import HybridLLMService
from voice_service import voice_service
from rag_service import search_government_docs
from policy_chart_generator import PolicyChartGenerator
from conversational_service import conversational_service

# Initialize hybrid LLM service
hybrid_llm_service = HybridLLMService()

logger = logging.getLogger("voicebot")

def analyze_query_context(query: str) -> dict:
    """Analyze query to determine if it's document-related or general, and identify user role"""
    query_lower = query.lower()
    
    # Role-specific keywords and queries
    role_patterns = {
        'pension_beneficiary': [
            'pension eligibility', 'pension documents', 'pension application', 'retirement benefits',
            'pension calculation', 'pension amount', 'family pension', 'commutation',
            'gratuity eligibility', 'provident fund withdrawal', 'medical benefits after retirement',
            'pension certificate', 'life certificate', 'pension arrears', 'how to apply pension',
            'pension office', 'pension disbursement', 'pension inquiry', 'pension status'
        ],
        'procurement_officer': [
            'tender process', 'bid submission', 'procurement thresholds', 'gem portal',
            'msme relaxation', 'vendor registration', 'procurement checklist', 'bid evaluation',
            'tender documents', 'procurement rules', 'bidding process', 'contract award',
            'procurement guidelines', 'tender notice', 'technical bid', 'financial bid',
            'procurement manual', 'vendor empanelment', 'tender committee'
        ],
        'finance_staff': [
            'sanctioning authority', 'financial approval', 'budget allocation', 'expenditure sanction',
            'financial registers', 'audit compliance', 'treasury rules', 'payment authorization',
            'financial delegation', 'budget utilization', 'fund release', 'financial procedure',
            'accounting rules', 'financial reporting', 'expenditure control', 'financial audit',
            'cash book', 'voucher processing', 'financial clearance'
        ],
        'leadership_policymaker': [
            'policy impact', 'scenario analysis', 'cost comparison', 'policy implementation',
            'evidence pack', 'policy evaluation', 'impact assessment', 'strategic planning',
            'policy formulation', 'comparative analysis', 'policy review', 'governance framework',
            'administrative reform', 'policy effectiveness', 'decision support', 'policy brief'
        ]
    }
    
    # Government document keywords (expanded)
    doc_keywords = [
        'pension', 'leave', 'allowance', 'da', 'dearness', 'procurement', 'tender',
        'medical', 'reimbursement', 'transfer', 'posting', 'promotion', 'service',
        'rules', 'policy', 'government', 'circular', 'notification', 'benefits',
        'gratuity', 'provident fund', 'retirement', 'salary', 'pay commission',
        'eligibility', 'documents', 'application', 'process', 'approval', 'sanction',
        'audit', 'finance', 'budget', 'expenditure', 'treasury', 'guidelines'
    ]
    
    # General conversation keywords
    general_keywords = [
        'hello', 'hi', 'thank you', 'thanks', 'goodbye', 'bye', 'help',
        'how are you', 'what is your name', 'who are you', 'weather',
        'time', 'date', 'joke', 'story', 'song', 'recipe', 'movie'
    ]
    
    # Detect user role
    detected_role = None
    role_confidence = 0.0
    
    for role, patterns in role_patterns.items():
        role_matches = sum(1 for pattern in patterns if pattern in query_lower)
        if role_matches > 0:
            current_confidence = min(role_matches * 0.4, 1.0)
            if current_confidence > role_confidence:
                detected_role = role
                role_confidence = current_confidence
    
    # Count general matches
    doc_matches = sum(1 for kw in doc_keywords if kw in query_lower)
    general_matches = sum(1 for kw in general_keywords if kw in query_lower)
    
    # Determine query type - FIXED: Be more aggressive about document searches
    if doc_matches > 0 or detected_role:
        query_type = "document_related"
        confidence = max(min(doc_matches * 0.3, 1.0), role_confidence)
    elif general_matches > 0 and doc_matches == 0:
        # Only treat as general if there are ZERO document keywords
        query_type = "general_conversation"
        confidence = min(general_matches * 0.4, 1.0)
    else:
        # DEFAULT to document search - this is a government document system
        query_type = "document_related"
        confidence = 0.5  # Higher confidence for document search by default
    
    return {
        "type": query_type,
        "confidence": confidence,
        "doc_keywords_found": doc_matches,
        "general_keywords_found": general_matches,
        "detected_role": detected_role,
        "role_confidence": role_confidence
    }

async def generate_llm_fallback_response(user_message: str, query_context: dict) -> str:
    """Generate response using Groq/Gemini for out-of-context queries"""
    try:
        # Determine which LLM to use based on query complexity
        provider = hybrid_llm_service.choose_llm_provider(user_message)
        
        # Create role-aware system prompt
        detected_role = query_context.get("detected_role")
        if query_context.get("type") == "general_conversation":
            system_prompt = """You are a helpful assistant for a government document system. 
            The user is asking a general question not related to government documents. 
            Provide a friendly, helpful response and gently guide them to ask about government policies, 
            pension rules, leave policies, procurement procedures, or other administrative matters if they need official information."""
        elif detected_role == "pension_beneficiary":
            system_prompt = """You are an AI assistant specializing in government pension and retirement benefits. 
            The user appears to be a pension beneficiary or claimant. Provide helpful information about pension eligibility, 
            application processes, required documents, and procedures. Always remind them to verify information with 
            the pension disbursing authority and consult official government sources for the most current rules."""
        elif detected_role == "procurement_officer":
            system_prompt = """You are an AI assistant specializing in government procurement procedures. 
            The user appears to be involved in procurement or bidding processes. Provide helpful information about 
            tender procedures, MSME benefits, GeM portal usage, and procurement guidelines. Always remind them to 
            follow current procurement rules and consult the latest government circulars."""
        elif detected_role == "finance_staff": 
            system_prompt = """You are an AI assistant specializing in government financial procedures. 
            The user appears to be finance staff. Provide helpful information about sanctioning procedures, 
            budget management, audit compliance, and treasury rules. Always remind them to follow current 
            financial rules and consult with the accounts department for official procedures."""
        elif detected_role == "leadership_policymaker":
            system_prompt = """You are an AI assistant specializing in policy analysis and decision support. 
            The user appears to be in a leadership or policy-making role. Provide helpful information about 
            policy impact analysis, evidence-based decision making, and strategic planning. Always recommend 
            consulting with relevant departments and conducting proper stakeholder consultations."""
        else:
            system_prompt = """You are an AI assistant for government document queries. 
            The user asked about something that wasn't found in the document database. 
            Provide helpful general information if you can, but always remind them that for official 
            government policies and procedures, they should consult official sources or contact 
            the relevant government office. Keep responses concise and professional."""
        
        # Generate response using hybrid LLM service
        if provider:
            response = await hybrid_llm_service.generate_response(
                user_message, 
                system_prompt=system_prompt,
                provider=provider
            )
            logger.info(f"βœ… Generated LLM fallback response using {provider.value}")
            return response
        else:
            logger.warning("⚠️ No LLM provider available")
            return "I understand your question, but I'm currently unable to access my AI capabilities. Please try again later or contact the relevant government office for official information."
            
    except Exception as e:
        logger.error(f"❌ Error generating LLM fallback response: {e}")
        return f"I apologize, but I encountered an error while processing your query: '{user_message}'. Please try rephrasing your question or contact the relevant authorities for assistance."

def validate_transcription_quality(text: str, language: str) -> dict:
    """Validate transcription quality and provide suggestions"""
    if not text or not text.strip():
        return {
            "score": 0.0,
            "level": "very_low",
            "suggestions": ["No speech detected", "Check microphone", "Speak closer to microphone"]
        }
    
    text_clean = text.strip()
    words = text_clean.split()
    
    # Quality indicators
    word_count = len(words)
    avg_word_length = sum(len(word) for word in words) / max(word_count, 1)
    has_meaningful_words = any(len(word) > 2 for word in words)
    
    # Check for garbled/nonsensical words (too many consonants, unusual patterns)
    garbled_words = 0
    for word in words:
        word_clean = ''.join(c for c in word.lower() if c.isalpha())
        if len(word_clean) > 3:
            consonants = sum(1 for c in word_clean if c not in 'aeiou')
            vowels = len(word_clean) - consonants
            if consonants > vowels * 2:  # Too many consonants
                garbled_words += 1
    
    garbled_ratio = garbled_words / max(word_count, 1)
    
    # Language-specific checks
    if language in ['en', 'hi-en']:
        # Check for common English/Hinglish patterns
        common_words = ['the', 'and', 'is', 'in', 'to', 'of', 'for', 'with', 'on', 'at', 'by', 'from',
                       'pension', 'government', 'policy', 'rules', 'what', 'how', 'why', 'when', 'where',
                       'benefits', 'allowance', 'service', 'employee', 'officer', 'department']
        has_common_words = any(word.lower() in common_words for word in words)
        
        # Check for obvious nonsensical combinations
        nonsensical_patterns = ['benchern', 'trend rules', 'rinterpret', 'wht']
        has_nonsensical = any(pattern in text_clean.lower() for pattern in nonsensical_patterns)
    else:
        has_common_words = True  # Assume valid for other languages
        has_nonsensical = False
    
    # Calculate quality score
    score = 0.0
    if word_count > 0:
        score += 0.2
    if word_count >= 3:
        score += 0.2
    if avg_word_length > 2:
        score += 0.2
    if has_meaningful_words:
        score += 0.2
    if has_common_words:
        score += 0.2
    
    # Apply penalties
    if garbled_ratio > 0.3:  # More than 30% garbled words
        score *= 0.3
    elif garbled_ratio > 0.1:  # More than 10% garbled words
        score *= 0.6
    
    if has_nonsensical:
        score *= 0.2
    
    if word_count < 2 or avg_word_length < 2:
        score *= 0.5
    
    # Determine quality level and suggestions
    if score >= 0.7:
        level = "high"
        suggestions = []
    elif score >= 0.4:
        level = "medium"
        suggestions = ["Speak a bit more clearly for better recognition"]
    elif score >= 0.2:
        level = "low"
        suggestions = ["Speak more clearly", "Try speaking slower", "Reduce background noise"]
    else:
        level = "very_low"
        suggestions = ["Audio quality is poor", "Speak closer to microphone", "Reduce background noise", "Try speaking more slowly and clearly"]
    
    return {
        "score": score,
        "level": level,
        "suggestions": suggestions,
        "garbled_ratio": garbled_ratio,
        "word_count": word_count
    }

def create_language_context(user_language: str, normalized_language: str) -> str:
    """Create appropriate language context for LLM responses"""
    if not user_language:
        return ""
    
    lang_lower = user_language.lower()
    
    if lang_lower in ['hindi', 'hi', 'hi-in']:
        return " (User is speaking in Hindi. You may include relevant Hindi terms for government policies in India, especially for technical terms like 'ΰ€Έΰ€°ΰ€•ΰ€Ύΰ€°ΰ₯€ ΰ€¨ΰ₯€ΰ€€ΰ€Ώ', 'ΰ€ͺΰ₯‡ΰ€‚ΰ€Άΰ€¨', 'ΰ€­ΰ€€ΰ₯ΰ€€ΰ€Ύ' etc.)"
    elif lang_lower in ['hinglish', 'hi-en']:
        return " (User is speaking in Hinglish - Hindi-English mix. Feel free to use both languages naturally in your response, especially for government terminology.)"
    elif lang_lower in ['spanish', 'es']:
        return " (User is speaking in Spanish. Respond in Spanish if possible, or provide translations for key terms.)"
    elif lang_lower in ['french', 'fr']:
        return " (User is speaking in French. Respond in French if possible, or provide translations for key terms.)"
    elif lang_lower in ['arabic', 'ar']:
        return " (User is speaking in Arabic. Respond in Arabic if possible, or provide translations for key terms.)"
    elif lang_lower in ['chinese', 'zh']:
        return " (User is speaking in Chinese. Respond in Chinese if possible, or provide translations for key terms.)"
    elif lang_lower in ['japanese', 'ja']:
        return " (User is speaking in Japanese. Respond in Japanese if possible, or provide translations for key terms.)"
    elif lang_lower in ['english', 'en', 'en-us', 'en-in']:
        return " (User is speaking in English. Provide clear, professional responses.)"
    else:
        return f" (User language preference: {user_language}. Adapt response accordingly if possible.)"

def select_voice_for_language(user_language: str, preferred_voice: str = None) -> str:
    """Select appropriate TTS voice based on user's language"""
    if preferred_voice:
        return preferred_voice
    
    if not user_language:
        return "en-US-AriaNeural"  # Default
    
    lang_lower = user_language.lower()
    
    # Voice mapping for different languages
    voice_map = {
        'hindi': 'hi-IN-SwaraNeural',
        'hi': 'hi-IN-SwaraNeural',
        'hi-in': 'hi-IN-SwaraNeural',
        'hinglish': 'en-IN-NeerjaNeural',  # Indian English for Hinglish
        'hi-en': 'en-IN-NeerjaNeural',
        'english': 'en-US-AriaNeural',
        'en': 'en-US-AriaNeural',
        'en-us': 'en-US-AriaNeural',
        'en-in': 'en-IN-NeerjaNeural',
        'spanish': 'es-ES-ElviraNeural',
        'es': 'es-ES-ElviraNeural',
        'french': 'fr-FR-DeniseNeural',
        'fr': 'fr-FR-DeniseNeural',
        'german': 'de-DE-KatjaNeural',
        'de': 'de-DE-KatjaNeural',
        'portuguese': 'pt-BR-FranciscaNeural',
        'pt': 'pt-BR-FranciscaNeural',
        'italian': 'it-IT-ElsaNeural',
        'it': 'it-IT-ElsaNeural',
        'russian': 'ru-RU-SvetlanaNeural',
        'ru': 'ru-RU-SvetlanaNeural',
        'chinese': 'zh-CN-XiaoxiaoNeural',
        'zh': 'zh-CN-XiaoxiaoNeural',
        'japanese': 'ja-JP-NanamiNeural',
        'ja': 'ja-JP-NanamiNeural',
        'arabic': 'ar-SA-ZariyahNeural',
        'ar': 'ar-SA-ZariyahNeural'
    }
    
    return voice_map.get(lang_lower, 'en-US-AriaNeural')

def attempt_transcription_correction(text: str, quality_info: dict) -> str:
    """Attempt to correct common transcription errors, especially for government terms"""
    if not text or quality_info.get('score', 1) > 0.6:
        return text  # Don't correct if quality is already good
    
    text_lower = text.lower()
    corrected = text
    
    # Common government term corrections
    corrections = {
        # Pension-related corrections
        'tension': 'pension',
        'penshun': 'pension',
        'penshan': 'pension',
        'mention': 'pension',
        'bruised': 'rules',
        'bruce': 'rules',
        'brews': 'rules',
        'cruise': 'rules',
        
        # Policy-related corrections
        'policy': 'policy',  # Keep as is
        'polity': 'policy',
        'polly': 'policy',
        
        # Government-related corrections
        'government': 'government',  # Keep as is
        'goverment': 'government',
        'govermint': 'government',
        
        # Allowance corrections
        'allowens': 'allowance',
        'alowance': 'allowance',
        
        # Benefits corrections
        'benifits': 'benefits',
        'benefets': 'benefits',
        
        # Common words
        'wat': 'what',
        'wot': 'what',
        'wen': 'when',
        'were': 'where',
        'haw': 'how',
        'no': 'know',
        'noe': 'know'
    }
    
    # Split into words and correct each
    words = corrected.split()
    corrected_words = []
    
    for word in words:
        # Remove punctuation for matching
        clean_word = word.lower().strip('.,!?;:')
        
        # Check for corrections
        if clean_word in corrections and corrections[clean_word] != clean_word:
            # Preserve original capitalization pattern
            if word.isupper():
                corrected_word = corrections[clean_word].upper()
            elif word.istitle():
                corrected_word = corrections[clean_word].capitalize()
            else:
                corrected_word = corrections[clean_word]
            
            # Preserve punctuation
            punctuation = word[len(clean_word):] if len(word) > len(clean_word) else ''
            corrected_words.append(corrected_word + punctuation)
        else:
            corrected_words.append(word)
    
    final_corrected = ' '.join(corrected_words)
    
    # Only return correction if it's significantly different
    if final_corrected.lower() != text.lower():
        return final_corrected
    
    return text

async def handle_enhanced_websocket_connection(websocket: WebSocket):
    """Enhanced WebSocket handler with hybrid LLM and voice features"""
    await websocket.accept()
    logger.info("πŸ”Œ Enhanced WebSocket client connected.")

    # Initialize session data
    session_data = {
        "messages": [],
        "user_preferences": {
            "voice_enabled": True,  # Enable voice by default since this is a voice bot
            "preferred_voice": "en-US-AriaNeural",
            "response_mode": "both"  # text, voice, both - default to both for voice bot
        },
        "context": ""
    }

    try:
        # Get initial connection data
        initial_data = await websocket.receive_json()
        
        # Validate initial data
        if not isinstance(initial_data, dict):
            logger.warning(f"⚠️ Invalid initial data format: {type(initial_data)}")
            initial_data = {}
        
        logger.info(f"πŸ“¨ Initial connection data: {initial_data}")
        
        # Extract user preferences
        if "preferences" in initial_data:
            session_data["user_preferences"].update(initial_data["preferences"])
        
        # Setup user session
        flag = "user_id" in initial_data
        graph = None  # Initialize graph variable
        
        if flag:
            thread_id = initial_data.get("user_id")
            knowledge_base = initial_data.get("knowledge_base", "government_docs")
            
            # Use hybrid LLM or traditional graph based on configuration
            if hybrid_llm_service.use_hybrid:
                logger.info("πŸ€– Using Hybrid LLM Service")
                use_hybrid = True
            else:
                graph = await create_graph(kb_tool=True, mcp_config=None)
                use_hybrid = False
            
            config = {
                "configurable": {
                    "thread_id": thread_id,
                    "knowledge_base": knowledge_base,
                }
            }
        else:
            # Basic setup for unauthenticated users
            thread_id = str(uuid.uuid4())
            knowledge_base = "government_docs"
            use_hybrid = hybrid_llm_service.use_hybrid
            
            if not use_hybrid:
                graph = create_basic_graph()
            
            config = {"configurable": {"thread_id": thread_id}}

        # Send initial greeting with voice/hybrid capabilities
        await send_enhanced_greeting(websocket, session_data)

        # Main message handling loop
        while True:
            try:
                data = await websocket.receive_json()
                
                # Validate message format
                if not isinstance(data, dict):
                    logger.warning(f"⚠️ Invalid message format: {type(data)}")
                    continue
                
                if "type" not in data:
                    logger.warning(f"⚠️ Message missing 'type' field: {data}")
                    continue
                
                message_type = data["type"]
                logger.debug(f"πŸ“¨ Received message type: {message_type}")
                
                if message_type == "text_message":
                    await handle_text_message(
                        websocket, data, session_data, 
                        use_hybrid, config, knowledge_base, graph
                    )
                
                elif message_type == "voice_message":
                    await handle_voice_message(
                        websocket, data, session_data,
                        use_hybrid, config, knowledge_base, graph
                    )
                
                elif message_type == "preferences_update":
                    await handle_preferences_update(websocket, data, session_data)
                
                elif message_type == "get_voice_status":
                    await websocket.send_json({
                        "type": "voice_status",
                        "data": voice_service.get_voice_status()
                    })
                
                elif message_type == "get_llm_status":
                    await websocket.send_json({
                        "type": "llm_status", 
                        "data": hybrid_llm_service.get_provider_info()
                    })
                
                elif message_type == "connection":
                    # Handle initial connection - already processed above
                    logger.debug("πŸ“¨ Connection message received (already processed)")
                
                elif message_type == "get_knowledge_bases":
                    # Handle knowledge base request
                    await websocket.send_json({
                        "type": "knowledge_bases",
                        "knowledge_bases": ["government_docs", "rajasthan_documents"]
                    })
                
                else:
                    logger.warning(f"⚠️ Unknown message type: {message_type}")

            except WebSocketDisconnect:
                logger.info("πŸ”Œ WebSocket client disconnected.")
                break
            except Exception as e:
                logger.error(f"❌ Error handling message: {e}")
                try:
                    await websocket.send_json({
                        "type": "error",
                        "message": f"An error occurred: {str(e)}"
                    })
                except:
                    pass  # Connection might be closed
            except Exception as e:
                logger.error(f"❌ Error handling message: {e}")
                await websocket.send_json({
                    "type": "error",
                    "message": f"An error occurred: {str(e)}"
                })

    except WebSocketDisconnect:
        logger.info("πŸ”Œ WebSocket client disconnected during setup.")
    except Exception as e:
        logger.error(f"❌ WebSocket error: {e}")
        try:
            await websocket.send_json({
                "type": "error",
                "message": f"Connection error: {str(e)}"
            })
        except:
            pass

async def send_enhanced_greeting(websocket: WebSocket, session_data: dict):
    """Send enhanced greeting with system capabilities"""
    
    # Get system status
    llm_info = hybrid_llm_service.get_provider_info()
    voice_status = voice_service.get_voice_status()
    
    greeting_text = f"""πŸ€– Welcome to the Government Document Assistant!

I'm powered by a hybrid AI system that can help you with:
β€’ Government policies and procedures
β€’ Document search and analysis 
β€’ Scenario analysis with visualizations
β€’ Quick answers and detailed explanations

Current capabilities:
β€’ LLM: {'Hybrid (' + llm_info['fast_provider'] + '/' + llm_info['complex_provider'] + ')' if llm_info['hybrid_enabled'] else 'Single provider'}
β€’ Voice features: {'Enabled' if voice_status['voice_enabled'] else 'Disabled'}

How can I assist you today? You can ask me about any government policies, procedures, or documents!"""

    # Send text greeting
    await websocket.send_json({
        "type": "connection_successful",
        "message": greeting_text,
        "provider_used": "system",
        "capabilities": {
            "hybrid_llm": llm_info['hybrid_enabled'],
            "voice_features": voice_status['voice_enabled'],
            "scenario_analysis": True
        }
    })

    # Send voice greeting if enabled
    if session_data["user_preferences"]["voice_enabled"] and voice_status['voice_enabled']:
        voice_greeting = "Welcome to the Government Document Assistant! I can help you with policies, procedures, and document analysis. How can I assist you today?"
        audio_data = await voice_service.text_to_speech(voice_greeting)
        
        if audio_data:
            await websocket.send_json({
                "type": "audio_response",
                "audio_data": base64.b64encode(audio_data).decode(),
                "format": "mp3"
            })

async def handle_text_message(websocket: WebSocket, data: dict, session_data: dict, 
                            use_hybrid: bool, config: dict, knowledge_base: str, graph=None):
    """Handle text message with hybrid LLM"""
    
    user_message = data["message"]
    logger.info(f"πŸ’¬ Received text message: {user_message}")
    
    # Send acknowledgment
    await websocket.send_json({
        "type": "message_received",
        "message": "Processing your message..."
    })

    try:
        if use_hybrid:
            # Stream hybrid LLM service response
            response_chunks = []
            provider_used = None
            async for chunk in get_hybrid_response(
                user_message, session_data["context"], config, knowledge_base, session_data.get("session_id")
            ):
                response_chunks.append(chunk)
                # Send each chunk as structured data
                await websocket.send_json({
                    "type": "streaming_response",
                    "clause_text": chunk.get("clause_text", ""),
                    "summary": chunk.get("summary", ""),
                    "role_checklist": chunk.get("role_checklist", []),
                    "source_title": chunk.get("source_title", ""),
                    "clause_id": chunk.get("clause_id", ""),
                    "date": chunk.get("date", ""),
                    "url": chunk.get("url", ""),
                    "score": chunk.get("score", 1.0),
                    "scenario_analysis": chunk.get("scenario_analysis", None),
                    "charts": chunk.get("charts", [])
                })
            # Optionally, aggregate or select the best chunk for final response
            # Here, just use the first chunk for context update and provider
            if response_chunks:
                provider_used = hybrid_llm_service.choose_llm_provider(user_message)
                provider_used = provider_used.value if provider_used else "unknown"
                session_data["context"] = response_chunks[0].get("clause_text", "")[-1000:]
        else:
            # Use traditional graph approach
            session_data["messages"].append(HumanMessage(content=user_message))
            result = await graph.ainvoke({"messages": session_data["messages"]}, config)
            response_text = result["messages"][-1].content
            provider_used = "traditional"
            await send_text_response(websocket, response_text, provider_used, session_data)

        await websocket.send_json({
            "type": "llm_response",
            "text": "Done",
            "provider_used": provider_used,
            "timestamp": asyncio.get_event_loop().time()
        })

    except Exception as e:
        logger.error(f"❌ Error processing text message: {e}")
        await websocket.send_json({
            "type": "error", 
            "message": f"Error processing your message: {str(e)}"
        })

async def handle_voice_message(websocket: WebSocket, data: dict, session_data: dict,
                             use_hybrid: bool, config: dict, knowledge_base: str, graph=None):
    """Handle voice message with enhanced multi-language ASR and TTS"""
    
    if not voice_service.is_voice_enabled():
        await websocket.send_json({
            "type": "error",
            "message": "Voice features are not enabled"
        })
        return

    try:
        # Get audio data - handle both old and new format
        if "audio_data" in data:
            audio_data = base64.b64decode(data["audio_data"])
        else:
            # Handle old format or direct binary data
            logger.error("❌ No audio_data field found in voice message")
            await websocket.send_json({
                "type": "error",
                "message": "No audio data provided"
            })
            return
        
        # Extract and validate user language preference
        user_language = data.get("lang") or data.get("language") or session_data.get("language") or session_data["user_preferences"].get("language") or "english"
        
        # Normalize language codes
        language_map = {
            'english': 'en', 'en': 'en', 'en-us': 'en', 'en-in': 'en',
            'hindi': 'hi', 'hi': 'hi', 'hi-in': 'hi',
            'hinglish': 'hi-en', 'hi-en': 'hi-en',
            'spanish': 'es', 'es': 'es',
            'french': 'fr', 'fr': 'fr',
            'german': 'de', 'de': 'de',
            'portuguese': 'pt', 'pt': 'pt',
            'italian': 'it', 'it': 'it',
            'russian': 'ru', 'ru': 'ru',
            'chinese': 'zh', 'zh': 'zh',
            'japanese': 'ja', 'ja': 'ja',
            'arabic': 'ar', 'ar': 'ar'
        }
        
        normalized_language = language_map.get(user_language.lower(), 'en')
        logger.info(f"🌍 Processing voice with language: {user_language} (normalized: {normalized_language})")
        
        # Save to temporary file
        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_file:
            temp_file.write(audio_data)
            temp_file_path = temp_file.name

        # Check if we should use server-side ASR or expect browser transcription
        if voice_service.asr_provider == "browser-native":
            # Expect transcription to come from browser, not from audio processing
            logger.info("οΏ½ Using browser-native ASR - expecting transcription from client")
            
            # Clean up temp file since we won't process it
            Path(temp_file_path).unlink()
            
            # Check if transcription was provided in the message
            if "transcription" in data:
                transcribed_text = data["transcription"]
                logger.info(f"🎀 Browser transcription ({user_language}): {transcribed_text}")
            else:
                await websocket.send_json({
                    "type": "info",
                    "message": "Browser ASR mode - please ensure your browser supports speech recognition"
                })
                return
        else:
            # Use server-side ASR (Whisper) with multiple attempts if needed
            logger.info(f"🎀 Processing audio with language preference: {user_language}")
            transcribed_text = await voice_service.speech_to_text(temp_file_path, normalized_language)
            
            # If transcription seems poor, try with English as fallback
            if transcribed_text and normalized_language != 'en':
                quality_check = validate_transcription_quality(transcribed_text, normalized_language)
                if quality_check['score'] < 0.3:
                    logger.info("πŸ”„ Trying English transcription as fallback")
                    english_transcription = await voice_service.speech_to_text(temp_file_path, 'en')
                    if english_transcription:
                        english_quality = validate_transcription_quality(english_transcription, 'en')
                        if english_quality['score'] > quality_check['score'] + 0.2:
                            logger.info(f"🎯 English transcription better: {english_transcription}")
                            transcribed_text = english_transcription
                            normalized_language = 'en'
            
            # Clean up temp file
            Path(temp_file_path).unlink()

            if not transcribed_text:
                await websocket.send_json({
                    "type": "error", 
                    "message": "Could not transcribe audio. Please try speaking clearly or check your microphone."
                })
                return

        # Validate and potentially correct transcription
        transcription_quality = validate_transcription_quality(transcribed_text, normalized_language)
        corrected_text = attempt_transcription_correction(transcribed_text, transcription_quality)
        
        # Use corrected text if available and quality improved
        final_text = corrected_text if corrected_text != transcribed_text else transcribed_text
        final_quality = validate_transcription_quality(final_text, normalized_language) if corrected_text != transcribed_text else transcription_quality
        
        logger.info(f"🎀 Transcribed ({user_language}): {transcribed_text} | Quality: {transcription_quality['score']:.2f}")
        if corrected_text != transcribed_text:
            logger.info(f"πŸ”§ Corrected to: {final_text} | New Quality: {final_quality['score']:.2f}")

        # Send transcription with quality info
        await websocket.send_json({
            "type": "transcription",
            "text": final_text,
            "original_text": transcribed_text if corrected_text != transcribed_text else None,
            "language": user_language or "auto-detected",
            "confidence": final_quality['level'],
            "quality_score": final_quality['score'],
            "suggestions": final_quality['suggestions'],
            "was_corrected": corrected_text != transcribed_text
        })

        # Handle low-quality transcription with detailed feedback
        if final_quality['score'] < 0.2:
            await websocket.send_json({
                "type": "transcription_error",
                "message": f"Could not understand the audio clearly. Transcribed: '{final_text}'. Please try again with clearer speech.",
                "suggestions": final_quality['suggestions'],
                "quality_details": {
                    "score": final_quality['score'],
                    "garbled_ratio": final_quality.get('garbled_ratio', 0),
                    "word_count": final_quality.get('word_count', 0)
                }
            })
            return
        elif final_quality['score'] < 0.4:
            # Continue processing but warn user
            correction_note = f" (Auto-corrected from: '{transcribed_text}')" if corrected_text != transcribed_text else ""
            await websocket.send_json({
                "type": "transcription_warning", 
                "message": f"Audio quality is low (Score: {final_quality['score']:.2f}). I heard: '{final_text}'{correction_note}. Is this correct?",
                "suggestions": final_quality['suggestions'] + ["Try speaking more slowly", "Ensure microphone is close to your mouth", "Reduce background noise"]
            })

        # Add comprehensive language context to the prompt for better responses
        language_context = create_language_context(user_language, normalized_language)
        enhanced_message = final_text + language_context

        # Process as text message with language context
        if use_hybrid:
            response_chunks = []
            async for chunk in get_hybrid_response(
                enhanced_message, session_data["context"], config, knowledge_base, session_data.get("session_id")
            ):
                response_chunks.append(chunk)
                # Send each chunk as structured data
                await websocket.send_json({
                    "type": "streaming_response",
                    "clause_text": chunk.get("clause_text", ""),
                    "summary": chunk.get("summary", ""),
                    "role_checklist": chunk.get("role_checklist", []),
                    "source_title": chunk.get("source_title", ""),
                    "clause_id": chunk.get("clause_id", ""),
                    "date": chunk.get("date", ""),
                    "url": chunk.get("url", ""),
                    "score": chunk.get("score", 1.0),
                    "scenario_analysis": chunk.get("scenario_analysis", None),
                    "charts": chunk.get("charts", [])
                })
            
            # Create response text for voice synthesis from the chunks
            response_text_parts = []
            for chunk in response_chunks:
                if chunk.get("clause_text"):
                    response_text_parts.append(chunk.get("clause_text", ""))
                if chunk.get("summary"):
                    response_text_parts.append(chunk.get("summary", ""))
            response_text = " ".join(response_text_parts) if response_text_parts else "I found relevant information about your query."
            
            provider_used = hybrid_llm_service.choose_llm_provider(enhanced_message)
            provider_used = provider_used.value if provider_used else "unknown"
        else:
            session_data["messages"].append(HumanMessage(content=enhanced_message))
            result = await graph.ainvoke({"messages": session_data["messages"]}, config)
            response_text = result["messages"][-1].content
            provider_used = "traditional"

        # Send text response
        await send_text_response(websocket, response_text, provider_used, session_data)

        # Send voice response if enabled
        if session_data["user_preferences"]["response_mode"] in ["voice", "both"]:
            # Choose appropriate voice based on user's language
            voice_preference = select_voice_for_language(user_language, session_data["user_preferences"]["preferred_voice"])

            voice_text = voice_service.create_voice_response_with_guidance(
                response_text,
                suggested_resources=["Government portal", "Local offices", "Helpline numbers"],
                redirect_info="contact your local government office for personalized assistance"
            )
            
            audio_response = await voice_service.text_to_speech(
                voice_text, 
                voice_preference
            )
            
            if audio_response:
                await websocket.send_bytes(audio_response)
            else:
                logger.warning("⚠️ Could not generate audio response")

    except Exception as e:
        logger.error(f"❌ Error processing voice message: {e}")
        await websocket.send_json({
            "type": "error",
            "message": f"Error processing voice message: {str(e)}. Please try again or switch to text mode."
        })

async def get_hybrid_response(user_message: str, context: str, config: dict, knowledge_base: str, session_id: str = None):
    """Get response using hybrid LLM with conversational clarity checks and intelligent document search"""
    try:
        # First, determine if this is a government document query or general query
        query_context = analyze_query_context(user_message)
        logger.info(f"πŸ” Query analysis: {query_context}")
        
        # Check for follow-up context from previous clarification requests
        follow_up_context = conversational_service.handle_follow_up(user_message, session_id) if session_id else {'is_follow_up': False}
        
        if follow_up_context['is_follow_up']:
            # Use enhanced query from follow-up context
            search_query = follow_up_context['enhanced_query']
            logger.info(f"οΏ½ Using follow-up enhanced query: '{search_query}'")
        else:
            search_query = user_message
        
        logger.info(f"οΏ½πŸ” Searching documents for: '{search_query}' in knowledge base: {knowledge_base}")
        from rag_service import search_documents_async
        docs = await search_documents_async(search_query, limit=5)  # Increased limit for better results
        logger.info(f"πŸ“Š Document search returned {len(docs) if docs else 0} results")
        
        # Conversational clarity analysis - check if we need clarification
        if not follow_up_context['is_follow_up']:  # Don't ask for clarification on follow-ups
            conversational_analysis = conversational_service.generate_conversational_response(
                user_message, docs, session_id
            )
            
            if conversational_analysis['needs_clarification']:
                logger.info("❓ Query needs clarification - asking user for more context")
                # Return clarification request instead of search results
                yield {
                    "clause_text": conversational_analysis['response'],
                    "summary": "Clarification request to better understand your question",
                    "role_checklist": ["Please provide more specific information"],
                    "source_title": "Conversational Assistant",
                    "clause_id": "CLARIFICATION_REQUEST",
                    "date": "2024",
                    "url": "",
                    "score": 1.0,
                    "scenario_analysis": None,
                    "charts": [],
                    "needs_clarification": True,
                    "query_type": conversational_analysis.get('query_type', 'unclear')
                }
                return
        
        # Check if we have relevant documents
        has_relevant_docs = docs and any(doc.get("score", 0) > 0.5 for doc in docs)
        
        # FIXED: Always try document search first, even for apparent "general" queries
        # This is a government document system - most queries should check documents
        # Only use pure LLM for very clear greetings/thanks with NO document matches
        very_general_keywords = ['hello', 'hi', 'thank you', 'thanks', 'goodbye', 'bye']
        is_very_general = (query_context.get("type") == "general_conversation" and 
                          query_context.get("confidence", 0) > 0.8 and
                          any(keyword in user_message.lower() for keyword in very_general_keywords) and
                          not docs)
        
        if is_very_general:
            logger.info("πŸ“± Detected pure greeting/thanks with no documents, using LLM directly")
            llm_response = await generate_llm_fallback_response(user_message, query_context)
            yield {
                "clause_text": llm_response,
                "summary": "AI-generated response for general conversation",
                "role_checklist": ["This is general information", "For official queries, ask about government policies"],
                "source_title": "AI Assistant",
                "clause_id": "AI_GENERAL",
                "date": "2024",
                "url": "",
                "score": 0.9,
                "scenario_analysis": None,
                "charts": []
            }
            return
        
        if has_relevant_docs:
            try:
                from scenario_analysis_service import run_scenario_analysis
                # Detect scenario analysis intent (simple keyword match)
                scenario_keywords = ["impact", "cost", "scenario", "multiplier", "da", "dr"]
                if any(kw in user_message.lower() for kw in scenario_keywords):
                    logger.info("πŸ” Running scenario analysis")
                    # Example params extraction (can be improved)
                    params = {
                        'base_pension': 30000,
                        'multiplier': 1.1 if "multiplier" in user_message.lower() else 1.0,
                        'da_percent': 0.06 if "da" in user_message.lower() else 0.0,
                        'num_beneficiaries': 1000,
                        'years': 3,
                        'inflation': 0.05
                    }
                    scenario_result = run_scenario_analysis(params)
                    
                    # Generate charts for scenario_result
                    try:
                        chart_gen = PolicyChartGenerator()
                        charts = []
                        # Example: line chart for yearly results
                        if "yearly_results" in scenario_result:
                            years = [r['year'] for r in scenario_result['yearly_results']]
                            base_costs = [r['base_cost'] for r in scenario_result['yearly_results']]
                            scenario_costs = [r['scenario_cost'] for r in scenario_result['yearly_results']]
                            # Generate chart and append to charts list
                            fig, ax = plt.subplots(figsize=(10, 6))
                            ax.plot(years, base_costs, label='Base Cost', marker='o')
                            ax.plot(years, scenario_costs, label='Scenario Cost', marker='s')
                            ax.legend()
                            ax.set_title('Scenario Analysis: Cost Over Years')
                            ax.set_xlabel('Year')
                            ax.set_ylabel('Cost (β‚Ή)')
                            ax.grid(True, alpha=0.3)
                            
                            # Format y-axis to show values in lakhs
                            ax.yaxis.set_major_formatter(plt.FuncFormatter(lambda x, p: f'β‚Ή{x/100000:.1f}L'))
                            
                            buf = io.BytesIO()
                            fig.savefig(buf, format='png', dpi=150, bbox_inches='tight')
                            buf.seek(0)
                            chart_base64 = base64.b64encode(buf.read()).decode('utf-8')
                            plt.close(fig)
                            charts.append({"type": "line_chart", "data": chart_base64})
                            logger.info(f"βœ… Generated {len(charts)} charts for scenario analysis")
                        scenario_result["charts"] = charts
                    except Exception as chart_error:
                        logger.error(f"❌ Failed to generate charts: {chart_error}")
                        scenario_result["charts"] = []
                        scenario_result["chart_error"] = str(chart_error)
                else:
                    scenario_result = None
            except Exception as scenario_error:
                logger.error(f"❌ Scenario analysis failed: {scenario_error}")
                scenario_result = None
            for doc in docs:
                response_obj = {
                    "clause_text": doc.get("clause_text", ""),
                    "summary": doc.get("summary", ""),
                    "role_checklist": doc.get("role_checklist", []),
                    "source_title": doc.get("source_title", ""),
                    "clause_id": doc.get("clause_id", ""),
                    "date": doc.get("date", ""),
                    "url": doc.get("url", ""),
                    "score": doc.get("score", 1.0),
                    "scenario_analysis": scenario_result,
                    "charts": scenario_result.get("charts", []) if scenario_result else []
                }
                yield response_obj
        else:
            # No relevant documents found - use LLM fallback
            logger.info("πŸ“š No relevant documents found, using LLM fallback")
            llm_response = await generate_llm_fallback_response(user_message, query_context)
            yield {
                "clause_text": llm_response,
                "summary": "Generated by AI assistant for general query",
                "role_checklist": ["Consider if this relates to government policies", "Contact relevant office for official information"],
                "source_title": "AI Assistant",
                "clause_id": "AI_001",
                "date": "2024",
                "url": "",
                "score": 0.8,
                "scenario_analysis": None,
                "charts": []
            }
    except Exception as e:
        logger.warning(f"❌ Document search failed: {e}, using LLM fallback")
        try:
            llm_response = await generate_llm_fallback_response(user_message, {"type": "unknown", "confidence": 0.3})
            yield {
                "clause_text": llm_response,
                "summary": "AI-generated response due to system error",
                "role_checklist": ["Verify information independently", "Try rephrasing your query"],
                "source_title": "AI Assistant (Fallback)",
                "clause_id": "AI_ERROR",
                "date": "2024",
                "url": "",
                "score": 0.5,
                "scenario_analysis": None,
                "charts": []
            }
        except Exception as fallback_error:
            logger.error(f"❌ LLM fallback also failed: {fallback_error}")
            yield {
                "clause_text": "I apologize, but I'm experiencing technical difficulties. Please try again later or rephrase your question.",
                "summary": "System error occurred",
                "role_checklist": ["Try again later", "Rephrase your question", "Contact technical support"],
                "source_title": "System Error",
                "clause_id": "ERROR_001",
                "date": "2024",
                "url": "",
                "score": 0.1,
                "scenario_analysis": None,
                "charts": []
            }

async def send_text_response(websocket: WebSocket, response_text: str, provider_used: str, session_data: dict):
    """Send text response to client"""
    
    await websocket.send_json({
        "type": "llm_response", 
        "text": response_text,
        "provider_used": provider_used,
        "timestamp": asyncio.get_event_loop().time()
    })
    
    # Update session context
    session_data["context"] = response_text[-1000:]  # Keep last 1000 chars as context

async def handle_scenario_response(websocket: WebSocket, response_text: str, provider_used: str):
    """Handle scenario analysis response with images"""
    
    parts = response_text.split("SCENARIO_ANALYSIS_IMAGE:")
    text_part = parts[0].strip()
    
    # Send text part
    if text_part:
        await websocket.send_json({
            "type": "llm_response",
            "text": text_part,
            "provider_used": provider_used
        })
    
    # Send image parts
    for i, part in enumerate(parts[1:], 1):
        try:
            image_data = part.strip()
            await websocket.send_json({
                "type": "scenario_image",
                "image_data": image_data,
                "image_index": i,
                "chart_type": "analysis"
            })
        except Exception as e:
            logger.error(f"Error sending scenario image {i}: {e}")

async def handle_preferences_update(websocket: WebSocket, data: dict, session_data: dict):
    """Handle user preferences update"""
    
    try:
        session_data["user_preferences"].update(data["preferences"])
        
        await websocket.send_json({
            "type": "preferences_updated",
            "preferences": session_data["user_preferences"]
        })
        
        logger.info(f"πŸ”§ Updated user preferences: {session_data['user_preferences']}")
        
    except Exception as e:
        logger.error(f"❌ Error updating preferences: {e}")
        await websocket.send_json({
            "type": "error",
            "message": f"Error updating preferences: {str(e)}"
        })

# Keep the original function for backward compatibility
async def handle_websocket_connection(websocket: WebSocket):
    """Original websocket handler for backward compatibility"""
    await handle_enhanced_websocket_connection(websocket)