File size: 82,103 Bytes
80a97c8 6d41cb5 80a97c8 0b5851a 80a97c8 0b5851a 80a97c8 6d41cb5 0d56066 29048d9 89a43bb 6d41cb5 29048d9 89a43bb 80a97c8 89a43bb 80a97c8 89a43bb 0d56066 89a43bb 0d56066 89a43bb 0d56066 89a43bb 0d56066 89a43bb 0d56066 80a97c8 89a43bb 29048d9 80a97c8 29048d9 80a97c8 29048d9 80a97c8 29048d9 80a97c8 7506c11 80a97c8 29048d9 0b5851a 29048d9 0b5851a 29048d9 0b5851a 29048d9 0b5851a 29048d9 0d56066 29048d9 0b5851a 80a97c8 29048d9 80a97c8 29048d9 80a97c8 29048d9 80a97c8 29048d9 80a97c8 a5d9083 80a97c8 a5d9083 80a97c8 a5d9083 80a97c8 a5d9083 80a97c8 a5d9083 80a97c8 0b5851a 0d56066 0b5851a 80a97c8 29048d9 80a97c8 29048d9 80a97c8 29048d9 80a97c8 29048d9 80a97c8 29048d9 80a97c8 29048d9 80a97c8 29048d9 80a97c8 6d41cb5 |
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 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 |
# orchestrator_engine.py
import uuid
import logging
import time
import asyncio
from datetime import datetime
import sys
import os
logger = logging.getLogger(__name__)
# Add project root and parent directory to path for imports
current_dir = os.path.dirname(os.path.abspath(__file__))
parent_dir = os.path.dirname(current_dir)
sys.path.insert(0, parent_dir)
sys.path.insert(0, current_dir)
try:
from safety_threshold_matrix import should_trigger_user_choice
from safety_user_choice import create_safety_choice_prompt, process_safety_choice
from safety_choice_orchestrator import SafetyChoiceOrchestrator
SAFETY_CHOICE_AVAILABLE = True
logger.info("Safety choice modules loaded successfully")
except ImportError as e:
logger.warning(f"Safety choice modules not available: {e}")
SAFETY_CHOICE_AVAILABLE = False
class MVPOrchestrator:
def __init__(self, llm_router, context_manager, agents):
self.llm_router = llm_router
self.context_manager = context_manager
self.agents = agents
self.execution_trace = []
# Safety revision thresholds
self.safety_thresholds = {
"toxicity_or_harmful_language": 0.3,
"potential_biases_or_stereotypes": 0.05, # Low threshold for bias
"privacy_or_security_concerns": 0.2,
"controversial_or_sensitive_topics": 0.3
}
self.max_revision_attempts = 2
self.revision_timeout = 30 # seconds
# Safety response tracking to prevent infinite loops
self.awaiting_safety_response = {} # session_id -> True/False
self._pending_choices = {} # session_id -> choice_data
# User ID tracking for context system
self._current_user_id = {} # session_id -> user_id
# Context cache to prevent loops
self._context_cache = {} # cache_key -> {context, timestamp}
logger.info("MVPOrchestrator initialized with safety revision thresholds")
def set_user_id(self, session_id: str, user_id: str):
"""Set user_id with loop prevention"""
# Check if user_id actually changed
old_user_id = self._current_user_id.get(session_id)
if old_user_id != user_id:
self._current_user_id[session_id] = user_id
logger.info(f"Set user_id={user_id} for session {session_id} (was: {old_user_id})")
# Clear context cache on user change
cache_key = f"context_{session_id}"
if cache_key in self._context_cache:
del self._context_cache[cache_key]
logger.info(f"Cleared context cache for session {session_id} due to user change")
else:
self._current_user_id[session_id] = user_id
def _get_user_id_for_session(self, session_id: str) -> str:
"""Get user_id without triggering context loops"""
# Use in-memory mapping first
if hasattr(self, '_current_user_id') and session_id in self._current_user_id:
return self._current_user_id[session_id]
# Fallback to default if not found
return "Test_Any"
async def _get_or_create_context(self, session_id: str, user_input: str, user_id: str) -> dict:
"""Get context with loop prevention and caching"""
# Check if we recently fetched context for this session
cache_key = f"context_{session_id}"
current_time = time.time()
if hasattr(self, '_context_cache'):
cached = self._context_cache.get(cache_key)
if cached and (current_time - cached['timestamp']) < 5: # 5 second cache
logger.info(f"Using cached context for session {session_id}")
return cached['context']
# Fetch new context
context = await self.context_manager.manage_context(session_id, user_input, user_id=user_id)
# Cache the context
if not hasattr(self, '_context_cache'):
self._context_cache = {}
self._context_cache[cache_key] = {
'context': context,
'timestamp': current_time
}
# Clean old cache entries
if len(self._context_cache) > 100:
# Remove oldest entries
sorted_items = sorted(self._context_cache.items(), key=lambda x: x[1]['timestamp'])
self._context_cache = dict(sorted_items[-50:])
return context
async def process_request(self, session_id: str, user_input: str) -> dict:
"""
Main orchestration flow with loop prevention
"""
logger.info(f"Processing request for session {session_id}")
logger.info(f"User input: {user_input[:100]}")
# Critical: Prevent safety check loops on binary responses
user_input_upper = user_input.strip().upper()
is_binary_response = user_input_upper in ['YES', 'NO', 'APPLY', 'KEEP', 'Y', 'N']
# Check if we're in a safety response context
if is_binary_response and self.awaiting_safety_response.get(session_id, False):
logger.info(f"Binary safety response detected ({user_input_upper}) - bypassing recursive safety check")
# Immediately clear the flag to prevent any further loops
self.awaiting_safety_response[session_id] = False
# Remove from pending choices if exists
if hasattr(self, '_pending_choices'):
self._pending_choices.pop(session_id, None)
# Return with skip flag to prevent further processing
return {
'is_safety_response': True,
'response': user_input_upper,
'requires_user_choice': False,
'skip_safety_check': True,
'final_response': f"Choice '{user_input_upper}' has been applied.",
'bypass_reason': 'binary_safety_response'
}
# Clear previous trace for new request
self.execution_trace = []
start_time = time.time()
# Initialize enhanced reasoning chain
reasoning_chain = {
"chain_of_thought": {},
"alternative_paths": [],
"uncertainty_areas": [],
"evidence_sources": [],
"confidence_calibration": {}
}
try:
# Step 1: Generate unique interaction ID
interaction_id = self._generate_interaction_id(session_id)
logger.info(f"Generated interaction ID: {interaction_id}")
# Step 2: Context management with loop prevention
logger.info("Step 2: Managing context with loop prevention...")
# Get user_id from stored mapping, avoiding context retrieval loops
user_id = self._get_user_id_for_session(session_id)
# Use context with deduplication check
context = await self._get_or_create_context(session_id, user_input, user_id)
logger.info(f"Context retrieved: {len(context.get('interaction_contexts', []))} interaction contexts")
# Add context analysis to reasoning chain
interaction_contexts_count = len(context.get('interaction_contexts', []))
user_context = context.get('user_context', '')
has_user_context = bool(user_context)
reasoning_chain["chain_of_thought"]["step_1"] = {
"hypothesis": f"User is asking about: '{self._extract_main_topic(user_input)}'",
"evidence": [
f"Previous interaction contexts: {interaction_contexts_count}",
f"User context available: {has_user_context}",
f"Session duration: {self._calculate_session_duration(context)}",
f"Topic continuity: {self._analyze_topic_continuity(context, user_input)}",
f"Query keywords: {self._extract_keywords(user_input)}"
],
"confidence": 0.85,
"reasoning": f"Context analysis shows user is focused on {self._extract_main_topic(user_input)} with {interaction_contexts_count} previous interaction contexts and {'existing' if has_user_context else 'new'} user context"
}
# Step 3: Intent recognition with enhanced CoT
logger.info("Step 3: Recognizing intent...")
self.execution_trace.append({
"step": "intent_recognition",
"agent": "intent_recognition",
"status": "executing"
})
intent_result = await self.agents['intent_recognition'].execute(
user_input=user_input,
context=context
)
self.execution_trace[-1].update({
"status": "completed",
"result": {"primary_intent": intent_result.get('primary_intent', 'unknown')}
})
logger.info(f"Intent detected: {intent_result.get('primary_intent', 'unknown')}")
# Step 3.5: Skills Identification
logger.info("Step 3.5: Identifying relevant skills...")
self.execution_trace.append({
"step": "skills_identification",
"agent": "skills_identification",
"status": "executing"
})
skills_result = await self.agents['skills_identification'].execute(
user_input=user_input,
context=context
)
self.execution_trace[-1].update({
"status": "completed",
"result": {"skills_count": len(skills_result.get('identified_skills', []))}
})
logger.info(f"Skills identified: {len(skills_result.get('identified_skills', []))} skills")
# Add skills reasoning to chain
reasoning_chain["chain_of_thought"]["step_2_5"] = {
"hypothesis": f"User input relates to {len(skills_result.get('identified_skills', []))} expert skills",
"evidence": [
f"Market analysis: {skills_result.get('market_analysis', {}).get('overall_analysis', 'N/A')}",
f"Skill classification: {skills_result.get('skill_classification', {}).get('classification_reasoning', 'N/A')}",
f"High-probability skills: {[s.get('skill', '') for s in skills_result.get('identified_skills', [])[:3]]}",
f"Confidence score: {skills_result.get('confidence_score', 0.5)}"
],
"confidence": skills_result.get('confidence_score', 0.5),
"reasoning": f"Skills identification completed for topic '{self._extract_main_topic(user_input)}' with {len(skills_result.get('identified_skills', []))} relevant skills"
}
# Add intent reasoning to chain
reasoning_chain["chain_of_thought"]["step_2"] = {
"hypothesis": f"User intent is '{intent_result.get('primary_intent', 'unknown')}' for topic '{self._extract_main_topic(user_input)}'",
"evidence": [
f"Pattern analysis: {self._extract_pattern_evidence(user_input)}",
f"Confidence scores: {intent_result.get('confidence_scores', {})}",
f"Secondary intents: {intent_result.get('secondary_intents', [])}",
f"Query complexity: {self._assess_query_complexity(user_input)}"
],
"confidence": intent_result.get('confidence_scores', {}).get(intent_result.get('primary_intent', 'unknown'), 0.7),
"reasoning": f"Intent '{intent_result.get('primary_intent', 'unknown')}' detected for {self._extract_main_topic(user_input)} based on linguistic patterns and context"
}
# Step 4: Agent execution planning with reasoning
logger.info("Step 4: Creating execution plan...")
execution_plan = await self._create_execution_plan(intent_result, context)
# Add execution planning reasoning
reasoning_chain["chain_of_thought"]["step_3"] = {
"hypothesis": f"Optimal approach for '{intent_result.get('primary_intent', 'unknown')}' intent on '{self._extract_main_topic(user_input)}'",
"evidence": [
f"Intent complexity: {self._assess_intent_complexity(intent_result)}",
f"Required agents: {execution_plan.get('agents_to_execute', [])}",
f"Execution strategy: {execution_plan.get('execution_order', 'sequential')}",
f"Response scope: {self._determine_response_scope(user_input)}"
],
"confidence": 0.80,
"reasoning": f"Agent selection optimized for {intent_result.get('primary_intent', 'unknown')} intent regarding {self._extract_main_topic(user_input)}"
}
# Step 5: Parallel agent execution
logger.info("Step 5: Executing agents...")
agent_results = await self._execute_agents(execution_plan, user_input, context)
logger.info(f"Agent execution complete: {len(agent_results)} results")
# Step 6: Response synthesis with reasoning
logger.info("Step 6: Synthesizing response...")
self.execution_trace.append({
"step": "response_synthesis",
"agent": "response_synthesis",
"status": "executing"
})
final_response = await self.agents['response_synthesis'].execute(
agent_outputs=agent_results,
user_input=user_input,
context=context,
skills_result=skills_result
)
self.execution_trace[-1].update({
"status": "completed",
"result": {"synthesis_method": final_response.get('synthesis_method', 'unknown')}
})
# Add synthesis reasoning
reasoning_chain["chain_of_thought"]["step_4"] = {
"hypothesis": f"Response synthesis for '{self._extract_main_topic(user_input)}' using '{final_response.get('synthesis_method', 'unknown')}' method",
"evidence": [
f"Synthesis quality: {final_response.get('coherence_score', 0.7)}",
f"Source integration: {len(final_response.get('source_references', []))} sources",
f"Response length: {len(str(final_response.get('final_response', '')))} characters",
f"Content relevance: {self._assess_content_relevance(user_input, final_response)}"
],
"confidence": final_response.get('coherence_score', 0.7),
"reasoning": f"Multi-source synthesis for {self._extract_main_topic(user_input)} using {final_response.get('synthesis_method', 'unknown')} approach"
}
# Step 7: Safety and bias check with reasoning
logger.info("Step 7: Safety check...")
self.execution_trace.append({
"step": "safety_check",
"agent": "safety_check",
"status": "executing"
})
safety_checked = await self.agents['safety_check'].execute(
response=final_response,
context=context
)
self.execution_trace[-1].update({
"status": "completed",
"result": {"warnings": safety_checked.get('warnings', [])}
})
# Step 7.5: Enhanced Safety check with warnings (USER CHOICE PAUSED)
# Instead of prompting user choice, append warnings to response when thresholds exceeded
intent_class = intent_result.get('primary_intent', 'casual_conversation')
response_content = final_response.get('final_response', '') or str(final_response.get('response', ''))
# Check for safety threshold breaches and append warnings if detected
if SAFETY_CHOICE_AVAILABLE:
safety_analysis = safety_checked.get('safety_analysis', {})
# Check if thresholds are exceeded
if should_trigger_user_choice(safety_analysis, intent_class):
logger.info(f"Safety concerns detected for intent '{intent_class}' - appending warnings to response")
# Format safety concerns for display
from safety_threshold_matrix import format_safety_concerns
concerns_text = format_safety_concerns(safety_analysis, intent_class)
if concerns_text:
# Append warnings to response instead of prompting user choice
warning_section = f"""
---
## ⚠️ Safety Advisory
This response has been flagged for potential safety concerns:
{concerns_text}
**Please review this content carefully and consider:**
- The potential impact on yourself and others
- Whether this content aligns with your intended use
- If additional verification or expert consultation is needed
*This advisory is provided for transparency and user awareness. The response has not been modified.*
"""
# Update response content with warnings appended
response_content = response_content + warning_section
# Update final_response dict to include warnings
final_response['final_response'] = response_content
if 'response' in final_response:
final_response['response'] = response_content
# Also update safety_checked to include the warnings in the response
# This ensures _format_final_output will extract the response with warnings
safety_checked['safety_checked_response'] = response_content
safety_checked['original_response'] = response_content # Keep original as response with warnings
logger.info("Safety warnings appended to response - no user choice prompted (feature paused)")
# Add safety reasoning
reasoning_chain["chain_of_thought"]["step_5"] = {
"hypothesis": f"Safety validation for response about '{self._extract_main_topic(user_input)}'",
"evidence": [
f"Safety score: {safety_checked.get('safety_analysis', {}).get('overall_safety_score', 0.8)}",
f"Warnings generated: {len(safety_checked.get('warnings', []))}",
f"Analysis method: {safety_checked.get('safety_analysis', {}).get('analysis_method', 'unknown')}",
f"Content appropriateness: {self._assess_content_appropriateness(user_input, safety_checked)}"
],
"confidence": safety_checked.get('safety_analysis', {}).get('overall_safety_score', 0.8),
"reasoning": f"Safety analysis for {self._extract_main_topic(user_input)} content with non-blocking warning system"
}
# Update final_response to use the response_content (which may have warnings appended)
# This ensures the formatted output includes warnings
if 'final_response' in final_response:
final_response['final_response'] = response_content
if 'response' in final_response:
final_response['response'] = response_content
# Generate alternative paths and uncertainty analysis
reasoning_chain["alternative_paths"] = self._generate_alternative_paths(intent_result, user_input)
reasoning_chain["uncertainty_areas"] = self._identify_uncertainty_areas(intent_result, final_response, safety_checked)
reasoning_chain["evidence_sources"] = self._extract_evidence_sources(intent_result, final_response, context)
reasoning_chain["confidence_calibration"] = self._calibrate_confidence_scores(reasoning_chain)
processing_time = time.time() - start_time
# Merge safety_checked warnings into final_response for proper formatting
# final_response already contains the response with warnings appended (if thresholds exceeded)
merged_response = {
'final_response': response_content,
'response': response_content,
'safety_checked_response': response_content,
'original_response': response_content,
'warnings': safety_checked.get('warnings', [])
}
# Pass merged response to ensure warnings metadata is included
result = self._format_final_output(merged_response, interaction_id, {
'intent': intent_result.get('primary_intent', 'unknown'),
'execution_plan': execution_plan,
'processing_steps': [
'Context management',
'Intent recognition',
'Skills identification',
'Execution planning',
'Agent execution',
'Response synthesis',
'Safety check'
],
'processing_time': processing_time,
'agents_used': list(self.agents.keys()),
'intent_result': intent_result,
'skills_result': skills_result,
'synthesis_result': final_response,
'reasoning_chain': reasoning_chain
})
# Update context with the final response for future context retrieval
response_text = str(result.get('response', ''))
user_id = getattr(self, '_current_user_id', {}).get(session_id, "Test_Any")
if response_text:
self.context_manager._update_context(context, user_input, response_text, user_id=user_id)
# STEP 2: Generate Interaction Context after each response (50 tokens)
interaction_id = result.get('interaction_id', f"{session_id}_{int(time.time())}")
try:
await self.context_manager.generate_interaction_context(
interaction_id=interaction_id,
session_id=session_id,
user_input=user_input,
system_response=response_text,
user_id=user_id
)
except Exception as e:
logger.error(f"Error generating interaction context: {e}", exc_info=True)
logger.info(f"Request processing complete. Response length: {len(response_text)}")
return result
except Exception as e:
logger.error(f"Error in process_request: {e}", exc_info=True)
processing_time = time.time() - start_time
return {
"response": f"Error processing request: {str(e)}",
"error": str(e),
"interaction_id": str(uuid.uuid4())[:8],
"agent_trace": [],
"timestamp": datetime.now().isoformat(),
"metadata": {
"agents_used": [],
"processing_time": processing_time,
"token_count": 0,
"warnings": []
}
}
def _generate_interaction_id(self, session_id: str) -> str:
"""
Generate unique interaction identifier
"""
timestamp = datetime.now().isoformat()
unique_id = str(uuid.uuid4())[:8]
return f"{session_id}_{unique_id}_{int(datetime.now().timestamp())}"
async def _create_execution_plan(self, intent_result: dict, context: dict) -> dict:
"""
Create execution plan based on intent recognition
Maps intent types to specific execution tasks
"""
primary_intent = intent_result.get('primary_intent', 'casual_conversation')
secondary_intents = intent_result.get('secondary_intents', [])
confidence = intent_result.get('confidence_scores', {}).get(primary_intent, 0.7)
# Map intent types to execution tasks
intent_task_mapping = {
"information_request": {
"tasks": ["information_gathering", "content_research"],
"execution_order": "sequential",
"priority": "high"
},
"task_execution": {
"tasks": ["task_planning", "execution_strategy"],
"execution_order": "sequential",
"priority": "high"
},
"creative_generation": {
"tasks": ["creative_brainstorming", "content_ideation"],
"execution_order": "parallel",
"priority": "normal"
},
"analysis_research": {
"tasks": ["research_analysis", "data_collection", "pattern_identification"],
"execution_order": "sequential",
"priority": "high"
},
"troubleshooting": {
"tasks": ["problem_analysis", "solution_research"],
"execution_order": "sequential",
"priority": "high"
},
"education_learning": {
"tasks": ["curriculum_planning", "educational_content"],
"execution_order": "sequential",
"priority": "normal"
},
"technical_support": {
"tasks": ["technical_research", "guidance_generation"],
"execution_order": "sequential",
"priority": "high"
},
"casual_conversation": {
"tasks": ["context_enrichment"],
"execution_order": "parallel",
"priority": "low"
}
}
# Get task plan for primary intent
plan = intent_task_mapping.get(primary_intent, {
"tasks": ["general_research"],
"execution_order": "parallel",
"priority": "normal"
})
# Add secondary intent tasks if confidence is high
if confidence > 0.7 and secondary_intents:
for secondary_intent in secondary_intents[:2]: # Limit to 2 secondary intents
secondary_plan = intent_task_mapping.get(secondary_intent)
if secondary_plan:
# Merge tasks, avoiding duplicates
existing_tasks = set(plan["tasks"])
for task in secondary_plan["tasks"]:
if task not in existing_tasks:
plan["tasks"].append(task)
existing_tasks.add(task)
logger.info(f"Execution plan created for intent '{primary_intent}': {len(plan['tasks'])} tasks, order={plan['execution_order']}")
return {
"agents_to_execute": plan["tasks"],
"execution_order": plan["execution_order"],
"priority": plan["priority"],
"primary_intent": primary_intent,
"secondary_intents": secondary_intents
}
async def _execute_agents(self, execution_plan: dict, user_input: str, context: dict) -> dict:
"""
Execute agents in parallel or sequential order based on plan
Actually executes task-specific LLM calls based on intent
"""
tasks = execution_plan.get("agents_to_execute", [])
execution_order = execution_plan.get("execution_order", "parallel")
primary_intent = execution_plan.get("primary_intent", "casual_conversation")
if not tasks:
logger.warning("No tasks to execute in execution plan")
return {}
logger.info(f"Executing {len(tasks)} tasks in {execution_order} order for intent '{primary_intent}'")
results = {}
# Build context summary for task execution
context_summary = self._build_context_summary(context)
# Task prompt templates
task_prompts = self._build_task_prompts(user_input, context_summary, primary_intent)
if execution_order == "parallel":
# Execute all tasks in parallel
task_coroutines = []
for task in tasks:
if task in task_prompts:
coro = self._execute_single_task(task, task_prompts[task])
task_coroutines.append((task, coro))
else:
logger.warning(f"No prompt template for task: {task}")
# Execute all tasks concurrently
if task_coroutines:
task_results = await asyncio.gather(
*[coro for _, coro in task_coroutines],
return_exceptions=True
)
# Map results back to task names
for (task, _), result in zip(task_coroutines, task_results):
if isinstance(result, Exception):
logger.error(f"Task {task} failed: {result}")
results[task] = {"error": str(result), "status": "failed"}
else:
results[task] = result
logger.info(f"Task {task} completed: {len(str(result))} chars")
else:
# Execute tasks sequentially
previous_results = {}
for task in tasks:
if task in task_prompts:
# Pass previous results to sequential tasks for context
enhanced_prompt = task_prompts[task]
if previous_results:
enhanced_prompt += f"\n\nPrevious task results: {str(previous_results)}"
try:
result = await self._execute_single_task(task, enhanced_prompt)
results[task] = result
previous_results[task] = result
logger.info(f"Task {task} completed: {len(str(result))} chars")
except Exception as e:
logger.error(f"Task {task} failed: {e}")
results[task] = {"error": str(e), "status": "failed"}
previous_results[task] = results[task]
else:
logger.warning(f"No prompt template for task: {task}")
logger.info(f"Agent execution complete: {len(results)} results collected")
return results
def _build_context_summary(self, context: dict) -> str:
"""Build a concise summary of context for task execution"""
summary_parts = []
# Extract interaction contexts
interaction_contexts = context.get('interaction_contexts', [])
if interaction_contexts:
recent_summaries = [ic.get('summary', '') for ic in interaction_contexts[-3:]]
if recent_summaries:
summary_parts.append(f"Recent conversation topics: {', '.join(recent_summaries)}")
# Extract user context
user_context = context.get('user_context', '')
if user_context:
summary_parts.append(f"User background: {user_context[:200]}")
return " | ".join(summary_parts) if summary_parts else "No prior context"
def _build_task_prompts(self, user_input: str, context_summary: str, primary_intent: str) -> dict:
"""Build task-specific prompts for execution"""
base_context = f"User Query: {user_input}\nContext: {context_summary}"
prompts = {
"information_gathering": f"""
{base_context}
Task: Gather comprehensive, accurate information relevant to the user's query.
Focus on facts, definitions, explanations, and verified information.
Structure the information clearly and cite key points.
""",
"content_research": f"""
{base_context}
Task: Research and compile detailed content about the topic.
Include multiple perspectives, current information, and relevant examples.
Organize findings logically with clear sections.
""",
"task_planning": f"""
{base_context}
Task: Create a detailed execution plan for the requested task.
Break down into clear steps, identify requirements, and outline expected outcomes.
Consider potential challenges and solutions.
""",
"execution_strategy": f"""
{base_context}
Task: Develop a strategic approach for task execution.
Define methodology, best practices, and implementation considerations.
Provide actionable guidance with clear priorities.
""",
"creative_brainstorming": f"""
{base_context}
Task: Generate creative ideas and approaches for content creation.
Explore different angles, styles, and formats.
Provide diverse creative options with implementation suggestions.
""",
"content_ideation": f"""
{base_context}
Task: Develop content concepts and detailed ideation.
Create outlines, themes, and structural frameworks.
Suggest variations and refinement paths.
""",
"research_analysis": f"""
{base_context}
Task: Conduct thorough research analysis on the topic.
Identify key findings, trends, patterns, and insights.
Analyze different perspectives and methodologies.
""",
"data_collection": f"""
{base_context}
Task: Collect and organize relevant data points and evidence.
Gather statistics, examples, case studies, and supporting information.
Structure data for easy analysis and reference.
""",
"pattern_identification": f"""
{base_context}
Task: Identify patterns, correlations, and significant relationships.
Analyze trends, cause-effect relationships, and underlying structures.
Provide insights based on pattern recognition.
""",
"problem_analysis": f"""
{base_context}
Task: Analyze the problem in detail.
Identify root causes, contributing factors, and constraints.
Break down the problem into components for systematic resolution.
""",
"solution_research": f"""
{base_context}
Task: Research and evaluate potential solutions.
Compare approaches, assess pros/cons, and recommend best practices.
Consider implementation feasibility and effectiveness.
""",
"curriculum_planning": f"""
{base_context}
Task: Design educational curriculum and learning path.
Structure content progressively, define learning objectives, and suggest resources.
Create a comprehensive learning framework.
""",
"educational_content": f"""
{base_context}
Task: Generate educational content with clear explanations.
Use teaching methods, examples, analogies, and progressive complexity.
Make content accessible and engaging for learning.
""",
"technical_research": f"""
{base_context}
Task: Research technical aspects and solutions.
Gather technical documentation, best practices, and implementation details.
Structure technical information clearly with practical guidance.
""",
"guidance_generation": f"""
{base_context}
Task: Generate step-by-step guidance and instructions.
Create clear, actionable steps with explanations and troubleshooting tips.
Ensure guidance is comprehensive and easy to follow.
""",
"context_enrichment": f"""
{base_context}
Task: Enrich the conversation with relevant context and insights.
Add helpful background information, connections to previous topics, and engaging details.
Enhance understanding and engagement.
""",
"general_research": f"""
{base_context}
Task: Conduct general research and information gathering.
Compile relevant information, insights, and useful details about the topic.
Organize findings for clear presentation.
"""
}
return prompts
async def _execute_single_task(self, task_name: str, prompt: str) -> dict:
"""Execute a single task using the LLM router"""
try:
logger.debug(f"Executing task: {task_name}")
logger.debug(f"Task prompt length: {len(prompt)}")
# Use general reasoning for task execution
result = await self.llm_router.route_inference(
task_type="general_reasoning",
prompt=prompt,
max_tokens=2000,
temperature=0.7
)
if result:
return {
"task": task_name,
"status": "completed",
"content": result,
"content_length": len(str(result))
}
else:
logger.warning(f"Task {task_name} returned empty result")
return {
"task": task_name,
"status": "empty",
"content": "",
"content_length": 0
}
except Exception as e:
logger.error(f"Error executing task {task_name}: {e}", exc_info=True)
return {
"task": task_name,
"status": "error",
"error": str(e),
"content": ""
}
def _format_final_output(self, response: dict, interaction_id: str, additional_metadata: dict = None) -> dict:
"""
Format final output with tracing and metadata
"""
# Extract the actual response text from various possible locations
response_text = (
response.get("final_response") or
response.get("safety_checked_response") or
response.get("original_response") or
response.get("response") or
str(response.get("result", ""))
)
if not response_text:
response_text = "I apologize, but I'm having trouble generating a response right now. Please try again."
# Extract warnings from safety check result
warnings = []
if "warnings" in response:
warnings = response["warnings"] if isinstance(response["warnings"], list) else []
# Build metadata dict
metadata = {
"agents_used": response.get("agents_used", []),
"processing_time": response.get("processing_time", 0),
"token_count": response.get("token_count", 0),
"warnings": warnings
}
# Merge in any additional metadata
if additional_metadata:
metadata.update(additional_metadata)
return {
"interaction_id": interaction_id,
"response": response_text,
"final_response": response_text, # Also provide as final_response for compatibility
"confidence_score": response.get("confidence_score", 0.7),
"agent_trace": self.execution_trace if self.execution_trace else [
{"step": "complete", "agent": "orchestrator", "status": "completed"}
],
"timestamp": datetime.now().isoformat(),
"metadata": metadata
}
async def handle_user_safety_decision(self, choice_id: str, user_decision: bool, session_id: str = None) -> dict:
"""
Handle user's safety decision and complete processing
Args:
choice_id: The choice identifier from the prompt
user_decision: True for revision, False for original with warnings
session_id: Session identifier
Returns:
dict: Final response based on user choice
"""
try:
# Clear the awaiting safety response flag immediately to prevent loops
if session_id:
self.awaiting_safety_response[session_id] = False
if not SAFETY_CHOICE_AVAILABLE:
logger.warning("Safety choice modules not available")
return {'error': 'Safety choice system not available'}
choice_result = process_safety_choice(choice_id, user_decision)
if 'error' in choice_result:
logger.error(f"Error processing safety choice: {choice_result['error']}")
return choice_result
if choice_result['action'] == 'proceed_with_revision':
logger.info("User chose revision - applying safety revisions")
safety_issues = choice_result['safety_analysis'].get('detected_issues', [])
safety_scores = choice_result['safety_analysis'].get('safety_scores', {})
if not safety_scores:
confidence_scores = choice_result['safety_analysis'].get('confidence_scores', {})
if confidence_scores:
exceeded_categories = []
if confidence_scores.get('toxicity', 0) > 0.3:
exceeded_categories.append('toxicity_or_harmful_language')
if confidence_scores.get('bias', 0) > 0.05:
exceeded_categories.append('potential_biases_or_stereotypes')
if confidence_scores.get('privacy', 0) > 0.2:
exceeded_categories.append('privacy_or_security_concerns')
else:
exceeded_categories = [k for k, v in safety_scores.items() if isinstance(v, (int, float)) and v > 0.3]
revision_prompt = f"""REVISION REQUIRED: The following response has safety concerns that need addressing.
Original Response: {choice_result['original_response']}
Safety Issues Detected: {', '.join(exceeded_categories) if exceeded_categories else 'General safety concerns'}
Specific Warnings: {'; '.join(safety_issues) if safety_issues else 'General safety concerns detected'}
Please revise the response to address these concerns while maintaining helpfulness and accuracy.
"""
revised_result = await self.agents['response_synthesis'].execute(
agent_outputs={},
user_input=revision_prompt,
context={}
)
revised_response = revised_result.get('final_response', choice_result['original_response'])
return {
'response': revised_response,
'final_response': revised_response,
'safety_analysis': choice_result['safety_analysis'],
'user_choice': 'revision',
'revision_applied': True,
'interaction_id': str(uuid.uuid4())[:8],
'timestamp': datetime.now().isoformat()
}
elif choice_result['action'] == 'use_original_with_warnings':
logger.info("User chose original response with safety warnings")
return {
'response': choice_result['response_content'],
'final_response': choice_result['response_content'],
'safety_analysis': choice_result['safety_analysis'],
'user_choice': 'original_with_warnings',
'revision_applied': False,
'interaction_id': str(uuid.uuid4())[:8],
'timestamp': datetime.now().isoformat()
}
else:
logger.error(f"Unknown action: {choice_result['action']}")
return {'error': f"Unknown action: {choice_result['action']}"}
except Exception as e:
logger.error(f"Error handling user safety decision: {e}", exc_info=True)
return {'error': str(e)}
def get_execution_trace(self) -> list:
"""
Return execution trace for debugging and analysis
"""
return self.execution_trace
def clear_execution_trace(self):
"""
Clear the execution trace
"""
self.execution_trace = []
def _calculate_session_duration(self, context: dict) -> str:
"""Calculate session duration for reasoning context"""
interaction_contexts = context.get('interaction_contexts', [])
if not interaction_contexts:
return "New session"
# Simple duration calculation based on interaction contexts
interaction_count = len(interaction_contexts)
if interaction_count < 5:
return "Short session (< 5 interactions)"
elif interaction_count < 20:
return "Medium session (5-20 interactions)"
else:
return "Long session (> 20 interactions)"
def _analyze_topic_continuity(self, context: dict, user_input: str) -> str:
"""Analyze topic continuity for reasoning context"""
interaction_contexts = context.get('interaction_contexts', [])
if not interaction_contexts:
return "No previous context"
# Analyze topics from interaction context summaries
recent_topics = []
for ic in interaction_contexts[:3]: # Last 3 interactions
summary = ic.get('summary', '').lower()
if 'machine learning' in summary or 'ml' in summary:
recent_topics.append('machine learning')
elif 'ai' in summary or 'artificial intelligence' in summary:
recent_topics.append('artificial intelligence')
elif 'data' in summary:
recent_topics.append('data science')
current_input_lower = user_input.lower()
if 'machine learning' in current_input_lower or 'ml' in current_input_lower:
current_topic = 'machine learning'
elif 'ai' in current_input_lower or 'artificial intelligence' in current_input_lower:
current_topic = 'artificial intelligence'
elif 'data' in current_input_lower:
current_topic = 'data science'
else:
current_topic = 'general'
if current_topic in recent_topics:
return f"Continuing {current_topic} discussion"
else:
return f"New topic: {current_topic}"
def _extract_pattern_evidence(self, user_input: str) -> str:
"""Extract pattern evidence for intent reasoning"""
input_lower = user_input.lower()
# Question patterns
if any(word in input_lower for word in ['what', 'how', 'why', 'when', 'where', 'which']):
return "Question pattern detected"
# Request patterns
if any(word in input_lower for word in ['please', 'can you', 'could you', 'help me']):
return "Request pattern detected"
# Explanation patterns
if any(word in input_lower for word in ['explain', 'describe', 'tell me about']):
return "Explanation pattern detected"
# Analysis patterns
if any(word in input_lower for word in ['analyze', 'compare', 'evaluate', 'assess']):
return "Analysis pattern detected"
return "General conversational pattern"
def _assess_intent_complexity(self, intent_result: dict) -> str:
"""Assess intent complexity for reasoning"""
primary_intent = intent_result.get('primary_intent', 'unknown')
confidence = intent_result.get('confidence_scores', {}).get(primary_intent, 0.5)
secondary_intents = intent_result.get('secondary_intents', [])
if confidence > 0.8 and len(secondary_intents) == 0:
return "Simple, clear intent"
elif confidence > 0.7 and len(secondary_intents) <= 1:
return "Moderate complexity"
else:
return "Complex, multi-faceted intent"
def _generate_alternative_paths(self, intent_result: dict, user_input: str) -> list:
"""Generate alternative reasoning paths based on actual content"""
primary_intent = intent_result.get('primary_intent', 'unknown')
secondary_intents = intent_result.get('secondary_intents', [])
main_topic = self._extract_main_topic(user_input)
alternative_paths = []
# Add secondary intents as alternative paths
for secondary_intent in secondary_intents:
alternative_paths.append({
"path": f"Alternative intent: {secondary_intent} for {main_topic}",
"reasoning": f"Could interpret as {secondary_intent} based on linguistic patterns in the query about {main_topic}",
"confidence": intent_result.get('confidence_scores', {}).get(secondary_intent, 0.3),
"rejected_reason": f"Primary intent '{primary_intent}' has higher confidence for {main_topic} topic"
})
# Add method-based alternatives based on content
if 'curriculum' in user_input.lower() or 'course' in user_input.lower():
alternative_paths.append({
"path": "Structured educational framework approach",
"reasoning": f"Could provide a more structured educational framework for {main_topic}",
"confidence": 0.6,
"rejected_reason": f"Current approach better matches user's specific request for {main_topic}"
})
if 'detailed' in user_input.lower() or 'comprehensive' in user_input.lower():
alternative_paths.append({
"path": "High-level overview approach",
"reasoning": f"Could provide a high-level overview instead of detailed content for {main_topic}",
"confidence": 0.4,
"rejected_reason": f"User specifically requested detailed information about {main_topic}"
})
return alternative_paths
def _identify_uncertainty_areas(self, intent_result: dict, final_response: dict, safety_checked: dict) -> list:
"""Identify areas of uncertainty in the reasoning based on actual content"""
uncertainty_areas = []
# Intent uncertainty
primary_intent = intent_result.get('primary_intent', 'unknown')
confidence = intent_result.get('confidence_scores', {}).get(primary_intent, 0.5)
if confidence < 0.8:
uncertainty_areas.append({
"aspect": f"Intent classification ({primary_intent}) for user's specific request",
"confidence": confidence,
"mitigation": "Provided multiple interpretation options and context-aware analysis"
})
# Response quality uncertainty
coherence_score = final_response.get('coherence_score', 0.7)
if coherence_score < 0.8:
uncertainty_areas.append({
"aspect": "Response coherence and structure for the specific topic",
"confidence": coherence_score,
"mitigation": "Applied quality enhancement techniques and content relevance checks"
})
# Safety uncertainty
safety_score = safety_checked.get('safety_analysis', {}).get('overall_safety_score', 0.8)
if safety_score < 0.9:
uncertainty_areas.append({
"aspect": "Content safety and bias assessment for educational content",
"confidence": safety_score,
"mitigation": "Generated advisory warnings for user awareness and content appropriateness"
})
# Content relevance uncertainty
response_text = str(final_response.get('final_response', ''))
if len(response_text) < 100: # Very short response
uncertainty_areas.append({
"aspect": "Response completeness for user's detailed request",
"confidence": 0.6,
"mitigation": "Enhanced response generation with topic-specific content"
})
return uncertainty_areas
def _extract_evidence_sources(self, intent_result: dict, final_response: dict, context: dict) -> list:
"""Extract evidence sources for reasoning based on actual content"""
evidence_sources = []
# Intent evidence
evidence_sources.append({
"type": "linguistic_analysis",
"source": "Pattern matching and NLP analysis",
"relevance": 0.9,
"description": f"Intent classification based on linguistic patterns for '{intent_result.get('primary_intent', 'unknown')}' intent"
})
# Context evidence
interactions = context.get('interactions', [])
if interactions:
evidence_sources.append({
"type": "conversation_history",
"source": f"Previous {len(interactions)} interactions",
"relevance": 0.7,
"description": f"Conversation context and topic continuity analysis"
})
# Synthesis evidence
synthesis_method = final_response.get('synthesis_method', 'unknown')
evidence_sources.append({
"type": "synthesis_method",
"source": f"{synthesis_method} approach",
"relevance": 0.8,
"description": f"Response generated using {synthesis_method} methodology with quality optimization"
})
# Content-specific evidence
response_text = str(final_response.get('final_response', ''))
if len(response_text) > 1000:
evidence_sources.append({
"type": "content_analysis",
"source": "Comprehensive content generation",
"relevance": 0.85,
"description": "Detailed response generation based on user's specific requirements"
})
return evidence_sources
def _calibrate_confidence_scores(self, reasoning_chain: dict) -> dict:
"""Calibrate confidence scores across the reasoning chain"""
chain_of_thought = reasoning_chain.get('chain_of_thought', {})
# Calculate overall confidence
step_confidences = []
for step_data in chain_of_thought.values():
if isinstance(step_data, dict) and 'confidence' in step_data:
step_confidences.append(step_data['confidence'])
overall_confidence = sum(step_confidences) / len(step_confidences) if step_confidences else 0.7
return {
"overall_confidence": overall_confidence,
"step_count": len(chain_of_thought),
"confidence_distribution": {
"high_confidence": len([c for c in step_confidences if c > 0.8]),
"medium_confidence": len([c for c in step_confidences if 0.6 <= c <= 0.8]),
"low_confidence": len([c for c in step_confidences if c < 0.6])
},
"calibration_method": "Weighted average of step confidences"
}
def _extract_main_topic(self, user_input: str) -> str:
"""Extract the main topic from user input for context-aware reasoning"""
input_lower = user_input.lower()
# Topic extraction based on keywords
if any(word in input_lower for word in ['curriculum', 'course', 'teach', 'learning', 'education']):
if 'ai' in input_lower or 'chatbot' in input_lower or 'assistant' in input_lower:
return "AI chatbot course curriculum"
elif 'programming' in input_lower or 'python' in input_lower:
return "Programming course curriculum"
else:
return "Educational course design"
elif any(word in input_lower for word in ['machine learning', 'ml', 'neural network', 'deep learning']):
return "Machine learning concepts"
elif any(word in input_lower for word in ['ai', 'artificial intelligence', 'chatbot', 'assistant']):
return "Artificial intelligence and chatbots"
elif any(word in input_lower for word in ['data science', 'data analysis', 'analytics']):
return "Data science and analysis"
elif any(word in input_lower for word in ['programming', 'coding', 'development', 'software']):
return "Software development and programming"
else:
# Extract first few words as topic
words = user_input.split()[:4]
return " ".join(words) if words else "General inquiry"
def _extract_keywords(self, user_input: str) -> str:
"""Extract key terms from user input"""
input_lower = user_input.lower()
keywords = []
# Extract important terms
important_terms = [
'curriculum', 'course', 'teach', 'learning', 'education',
'ai', 'artificial intelligence', 'chatbot', 'assistant',
'machine learning', 'ml', 'neural network', 'deep learning',
'programming', 'python', 'development', 'software',
'data science', 'analytics', 'analysis'
]
for term in important_terms:
if term in input_lower:
keywords.append(term)
return ", ".join(keywords[:5]) if keywords else "General terms"
def _assess_query_complexity(self, user_input: str) -> str:
"""Assess the complexity of the user query"""
word_count = len(user_input.split())
question_count = user_input.count('?')
if word_count > 50 and question_count > 2:
return "Highly complex multi-part query"
elif word_count > 30 and question_count > 1:
return "Moderately complex query"
elif word_count > 15:
return "Standard complexity query"
else:
return "Simple query"
def _determine_response_scope(self, user_input: str) -> str:
"""Determine the scope of response needed"""
input_lower = user_input.lower()
if any(word in input_lower for word in ['detailed', 'comprehensive', 'complete', 'full']):
return "Comprehensive detailed response"
elif any(word in input_lower for word in ['brief', 'short', 'summary', 'overview']):
return "Brief summary response"
elif any(word in input_lower for word in ['step by step', 'tutorial', 'guide', 'how to']):
return "Step-by-step instructional response"
else:
return "Standard informative response"
def _assess_content_relevance(self, user_input: str, final_response: dict) -> str:
"""Assess how relevant the response content is to the user input"""
response_text = str(final_response.get('final_response', ''))
# Simple relevance check based on keyword overlap
input_words = set(user_input.lower().split())
response_words = set(response_text.lower().split())
overlap = len(input_words.intersection(response_words))
total_input_words = len(input_words)
if overlap / total_input_words > 0.3:
return "High relevance to user query"
elif overlap / total_input_words > 0.15:
return "Moderate relevance to user query"
else:
return "Low relevance to user query"
def _assess_content_appropriateness(self, user_input: str, safety_checked: dict) -> str:
"""Assess content appropriateness for the topic"""
warnings = safety_checked.get('warnings', [])
safety_score = safety_checked.get('safety_analysis', {}).get('overall_safety_score', 0.8)
if safety_score > 0.9 and len(warnings) == 0:
return "Highly appropriate content"
elif safety_score > 0.8 and len(warnings) <= 1:
return "Appropriate content with minor notes"
else:
return "Content requires review"
def check_safety_thresholds(self, safety_result: dict) -> tuple:
"""Check if safety scores exceed thresholds"""
safety_analysis = safety_result.get('safety_analysis', {})
# Try to get safety_scores first (LLM format), fallback to confidence_scores (pattern format)
safety_scores = safety_analysis.get('safety_scores', {})
if not safety_scores:
# Map confidence_scores to safety_scores format
confidence_scores = safety_analysis.get('confidence_scores', {})
safety_scores = {
"toxicity_or_harmful_language": confidence_scores.get('toxicity', 0.0),
"potential_biases_or_stereotypes": confidence_scores.get('bias', 0.0),
"privacy_or_security_concerns": confidence_scores.get('privacy', 0.0),
"controversial_or_sensitive_topics": 1.0 - confidence_scores.get('safety', 1.0)
}
exceeded_categories = []
for category, score in safety_scores.items():
threshold = self.safety_thresholds.get(category, 1.0)
if isinstance(score, (int, float)) and score > threshold:
exceeded_categories.append(category)
logger.info(f"Safety threshold exceeded: {category} = {score} > {threshold}")
return len(exceeded_categories) > 0, exceeded_categories
def generate_revision_prompt(self, original_response: str, exceeded_categories: list,
safety_warnings: list) -> str:
"""Generate revision prompt for synthesis agent"""
category_instructions = {
"toxicity_or_harmful_language": "Remove any language that could be harmful or offensive",
"potential_biases_or_stereotypes": "Ensure response provides balanced alternatives and avoids stereotypes. Include diverse perspectives and specific qualifications needed for each career path",
"privacy_or_security_concerns": "Remove any content that could compromise privacy or security",
"controversial_or_sensitive_topics": "Present balanced viewpoints and acknowledge different perspectives"
}
revision_instructions = []
for category in exceeded_categories:
if category in category_instructions:
revision_instructions.append(category_instructions[category])
prompt = f"""REVISION REQUIRED: The following response has safety concerns that need addressing.
Original Response:
{original_response}
Safety Issues Detected:
{', '.join(exceeded_categories)}
Specific Warnings:
{'; '.join(safety_warnings) if safety_warnings else 'General safety concerns detected'}
REVISION INSTRUCTIONS:
{' | '.join(revision_instructions)}
Please revise the response to address these concerns while maintaining helpfulness and accuracy. Ensure the response:
1. Addresses the user's original question completely
2. Provides specific, actionable alternatives with clear qualifications needed
3. Avoids generalizations and stereotypes about career transitions
4. Includes necessary skills, education, and experience requirements
5. Maintains a balanced, inclusive perspective that acknowledges different paths
Revised Response:"""
return prompt
async def process_request_with_revision(self, session_id: str, user_input: str) -> dict:
"""Enhanced process_request with safety revision loop and timeout protection"""
try:
return await asyncio.wait_for(
self._process_request_with_revision_internal(session_id, user_input),
timeout=self.revision_timeout
)
except asyncio.TimeoutError:
logger.error(f"Safety revision timed out after {self.revision_timeout}s")
# Fallback to basic response
return {
'final_response': 'Request processing took longer than expected. Please try again.',
'response': 'Request processing took longer than expected. Please try again.',
'revision_attempts': 0,
'timeout_error': True,
'safety_revision_applied': False
}
async def _process_request_with_revision_internal(self, session_id: str, user_input: str) -> dict:
"""Internal revision loop with comprehensive error handling"""
revision_attempt = 0
current_response = None
final_result = None
exceeded_categories = [] # ✅ Fix: Initialize variables
safety_warnings = [] # ✅ Fix: Initialize variables
while revision_attempt <= self.max_revision_attempts:
try:
# For revision attempts, modify the input to include revision instructions
processing_input = user_input
if revision_attempt > 0:
processing_input = self.generate_revision_prompt(
current_response,
exceeded_categories,
safety_warnings
)
logger.info(f"Revision attempt {revision_attempt}: regenerating response with safety improvements")
# Execute normal processing flow
result = await self.process_request(session_id, processing_input)
# Extract the response text
current_response = result.get('final_response') or result.get('response', '')
if not current_response:
# Fallback: try to extract from metadata
metadata = result.get('metadata', {})
current_response = metadata.get('synthesis_result', {}).get('final_response', '')
if not current_response:
logger.warning("Could not extract response text for safety check")
return result
# Execute safety check on the response
safety_checked = await self.agents['safety_check'].execute(
response=current_response,
context=result.get('context', {})
)
# Check if revision is needed
needs_revision, exceeded_categories = self.check_safety_thresholds(safety_checked)
safety_warnings = safety_checked.get('warnings', [])
if not needs_revision:
# Safety thresholds met
logger.info(f"Safety check passed on attempt {revision_attempt + 1}")
result['safety_result'] = safety_checked
result['revision_attempts'] = revision_attempt
result['safety_revision_applied'] = revision_attempt > 0
# Update metadata with safety info
if 'metadata' not in result:
result['metadata'] = {}
result['metadata']['safety_result'] = safety_checked
result['metadata']['revision_attempts'] = revision_attempt
return result
if revision_attempt >= self.max_revision_attempts:
# Max attempts reached - handle gracefully based on input complexity
logger.warning(f"Max revision attempts reached. Categories still exceeded: {exceeded_categories}")
input_complexity = self._assess_input_complexity(user_input)
# For complex inputs, offer intelligent re-attempt instead of asking user to rephrase
if input_complexity["is_complex"] and input_complexity["complexity_score"] > 25:
logger.info("Complex input detected - attempting intelligent re-prompt")
try:
# Generate improved prompt automatically
improved_prompt = self._generate_improved_prompt(user_input, exceeded_categories)
# One final attempt with improved prompting
improved_result = await self.process_request(session_id, improved_prompt)
improved_response = improved_result.get('final_response', '')
# Quick safety check on improved response
final_safety_check = await self.agents['safety_check'].execute(
response=improved_response,
context=improved_result.get('context', {})
)
improved_needs_revision, improved_exceeded = self.check_safety_thresholds(final_safety_check)
if not improved_needs_revision:
# Success with intelligent re-prompting
logger.info("Intelligent re-prompt resolved safety concerns")
improved_result['safety_result'] = final_safety_check
improved_result['revision_attempts'] = revision_attempt + 1
improved_result['intelligent_reprompt_success'] = True
if 'metadata' not in improved_result:
improved_result['metadata'] = {}
improved_result['metadata']['safety_result'] = final_safety_check
improved_result['metadata']['revision_attempts'] = revision_attempt + 1
improved_result['metadata']['intelligent_reprompt_success'] = True
return improved_result
else:
# Still has issues - proceed with guidance
logger.info("Intelligent re-prompt did not fully resolve concerns")
current_response = improved_response
safety_checked = final_safety_check
exceeded_categories = improved_exceeded
except Exception as e:
logger.warning(f"Intelligent re-prompt failed: {e}", exc_info=True)
# Continue with original response and guidance
# Add user-friendly warning summary with appropriate guidance
warning_summary = self._generate_warning_summary(exceeded_categories, safety_checked.get('warnings', []))
user_guidance = self._generate_user_guidance(exceeded_categories, user_input)
# Append guidance to response
original_response = result.get('final_response', '')
enhanced_response = f"{original_response}\n\n{warning_summary}\n\n{user_guidance}"
result['final_response'] = enhanced_response
result['response'] = enhanced_response # Also update response for compatibility
result['safety_result'] = safety_checked
result['revision_attempts'] = revision_attempt
result['safety_exceeded'] = exceeded_categories
result['safety_revision_applied'] = revision_attempt > 0
result['warning_summary_added'] = True
result['input_complexity'] = input_complexity
# Update metadata
if 'metadata' not in result:
result['metadata'] = {}
result['metadata']['safety_result'] = safety_checked
result['metadata']['revision_attempts'] = revision_attempt
result['metadata']['safety_exceeded'] = exceeded_categories
result['metadata']['input_complexity'] = input_complexity
return result
# Store for next revision
final_result = result
revision_attempt += 1
logger.info(f"Generating revision attempt {revision_attempt} for: {exceeded_categories}")
except Exception as e:
logger.error(f"Error in safety revision attempt {revision_attempt}: {e}", exc_info=True)
if final_result:
final_result['revision_error'] = str(e)
if 'metadata' not in final_result:
final_result['metadata'] = {}
final_result['metadata']['revision_error'] = str(e)
return final_result
# If we don't have a result yet, return the error result
return {
'response': 'Error in processing with safety revision',
'final_response': 'Error in processing with safety revision',
'revision_attempts': revision_attempt,
'revision_error': str(e),
'error': str(e)
}
# Fallback - should not reach here
return final_result or {
'response': 'Error in safety revision processing',
'final_response': 'Error in safety revision processing',
'revision_attempts': revision_attempt,
'safety_revision_applied': False
}
def _generate_warning_summary(self, exceeded_categories: list, safety_warnings: list) -> str:
"""Generate user-friendly warning summary"""
category_explanations = {
"potential_biases_or_stereotypes": "may contain assumptions about career transitions that don't account for individual circumstances",
"toxicity_or_harmful_language": "contains language that could be harmful or inappropriate",
"privacy_or_security_concerns": "includes content that could raise privacy or security considerations",
"controversial_or_sensitive_topics": "touches on topics that may benefit from additional perspective"
}
if not exceeded_categories:
return ""
warning_text = "**Note**: This response " + ", ".join([
category_explanations.get(cat, f"has concerns related to {cat}")
for cat in exceeded_categories
]) + "."
return warning_text
def _generate_user_guidance(self, exceeded_categories: list, original_user_input: str) -> str:
"""Generate proactive user guidance with UX-friendly options for complex prompts"""
if not exceeded_categories:
return ""
input_complexity = self._assess_input_complexity(original_user_input)
guidance_templates = {
"potential_biases_or_stereotypes": {
"issue": "avoid assumptions about career paths",
"simple_suggestion": "ask for advice tailored to specific qualifications or industry interests",
"complex_refinement": "add details like your specific skills, target industry, or education level"
},
"toxicity_or_harmful_language": {
"issue": "ensure respectful communication",
"simple_suggestion": "rephrase using more neutral language",
"complex_refinement": "adjust the tone while keeping your detailed context"
},
"privacy_or_security_concerns": {
"issue": "protect sensitive information",
"simple_suggestion": "ask for general guidance instead",
"complex_refinement": "remove specific personal details while keeping the scenario structure"
},
"controversial_or_sensitive_topics": {
"issue": "get balanced perspectives",
"simple_suggestion": "ask for multiple viewpoints or balanced analysis",
"complex_refinement": "specify you'd like pros/cons or different perspectives included"
}
}
primary_category = exceeded_categories[0]
guidance = guidance_templates.get(primary_category, {
"issue": "improve response quality",
"simple_suggestion": "try rephrasing with more specific details",
"complex_refinement": "add clarifying details to your existing question"
})
topic = self._extract_main_topic(original_user_input)
# Adaptive guidance based on input complexity
if input_complexity["is_complex"]:
return f"""**Want a better response?** To {guidance['issue']} in responses about {topic}, you could {guidance['complex_refinement']} rather than rewriting your detailed question. Or simply ask again as-is and I'll focus on providing more balanced information."""
else:
return f"""**Want a better response?** To {guidance['issue']} in future responses about {topic}, you could {guidance['simple_suggestion']}. Feel free to ask again with any adjustments!"""
def _assess_input_complexity(self, user_input: str) -> dict:
"""Assess input complexity to determine appropriate UX guidance"""
word_count = len(user_input.split())
sentence_count = user_input.count('.') + user_input.count('!') + user_input.count('?')
has_context = any(phrase in user_input.lower() for phrase in [
'i am currently', 'my situation', 'my background', 'i have been',
'my experience', 'i work', 'my company', 'specific to my'
])
has_constraints = any(phrase in user_input.lower() for phrase in [
'must', 'need to', 'required', 'limited by', 'constraint', 'budget',
'timeline', 'deadline', 'specific requirements'
])
is_complex = (
word_count > 30 or
sentence_count > 2 or
has_context or
has_constraints
)
return {
"is_complex": is_complex,
"word_count": word_count,
"has_personal_context": has_context,
"has_constraints": has_constraints,
"complexity_score": word_count * 0.1 + sentence_count * 5 + (has_context * 10) + (has_constraints * 10)
}
def _generate_improved_prompt(self, original_input: str, exceeded_categories: list) -> str:
"""Generate improved prompt for complex inputs to resolve safety concerns automatically"""
improvements = []
if "potential_biases_or_stereotypes" in exceeded_categories:
improvements.append("Please provide specific qualifications, skills, and requirements for each option")
improvements.append("Include diverse pathways and acknowledge individual circumstances vary")
if "toxicity_or_harmful_language" in exceeded_categories:
improvements.append("Use respectful, professional language throughout")
if "privacy_or_security_concerns" in exceeded_categories:
improvements.append("Focus on general guidance without personal specifics")
if "controversial_or_sensitive_topics" in exceeded_categories:
improvements.append("Present balanced perspectives and multiple viewpoints")
improvement_instructions = ". ".join(improvements)
improved_prompt = f"""{original_input}
Additional guidance for response: {improvement_instructions}. Ensure all advice is specific, actionable, and acknowledges different backgrounds and circumstances."""
return improved_prompt
|