File size: 156,181 Bytes
cba5a66 af226a7 3c9f801 d6fab25 cba5a66 d6fab25 3123910 cba5a66 d6fab25 5a91767 d6fab25 cba5a66 5a91767 cba5a66 5a91767 cba5a66 d6fab25 cba5a66 52e8e7a 030303e cba5a66 dbe879e cba5a66 d6fab25 dbe879e 0e4a811 dbe879e 46b58d7 61c709c fd44c51 61c709c 850db0e fd44c51 850db0e 4b6d98d 370041e 4b6d98d eb59e59 628312d eb59e59 628312d eb59e59 628312d eb59e59 628312d eb59e59 628312d eb59e59 628312d eb59e59 628312d eb59e59 628312d eb59e59 628312d eb59e59 628312d eb59e59 567fa26 5c70b17 628312d 5c70b17 628312d 5c70b17 628312d 5c70b17 628312d 5c70b17 628312d 5c70b17 628312d 5c70b17 628312d 5c70b17 628312d 5c70b17 fde9594 5c70b17 628312d 5c70b17 eb59e59 567fa26 eb59e59 567fa26 eb59e59 567fa26 eb59e59 567fa26 eb59e59 567fa26 eb59e59 5c70b17 850db0e 61c709c 628312d c082084 628312d c082084 628312d c082084 61c709c dbe879e d6fab25 cba5a66 d6fab25 567fa26 d6fab25 5a91767 46b58d7 5a91767 d6fab25 e8ece8a 5a91767 52e8e7a 5a91767 d547805 5a91767 d6fab25 cba5a66 d6fab25 cba5a66 567fa26 cba5a66 567fa26 cba5a66 d6fab25 cba5a66 d6fab25 cba5a66 f20d300 d6fab25 f20d300 cba5a66 f20d300 cba5a66 f20d300 cba5a66 f20d300 cba5a66 f20d300 cba5a66 d6fab25 d0069f8 0e4a811 d0069f8 0e4a811 d0069f8 0e4a811 d0069f8 0e4a811 d0069f8 0e4a811 d0069f8 0e4a811 d0069f8 0e4a811 d0069f8 0e4a811 d0069f8 0e4a811 d0069f8 0e4a811 d0069f8 0e4a811 d0069f8 d6fab25 d0069f8 d6fab25 d0069f8 d6fab25 d0069f8 d6fab25 18d3583 d6fab25 18d3583 d6fab25 18d3583 05120d5 d6fab25 9b99c4f 87269c3 9b99c4f d6fab25 9b99c4f d6fab25 9b99c4f cba5a66 52e8e7a d6fab25 52e8e7a 87269c3 6f7b8a9 87269c3 6f7b8a9 87269c3 6f7b8a9 87269c3 6f7b8a9 d6fab25 52e8e7a d6fab25 852ee83 d6fab25 79de83d d6fab25 79de83d 61c709c 79de83d 7b86bd0 bc9bce5 c082084 bc9bce5 79de83d 61c709c e8ece8a 79de83d 50df459 d6fab25 50df459 d6fab25 52e8e7a d6fab25 bc9bce5 d6fab25 79de83d 50df459 79de83d 50df459 79de83d 50df459 79de83d 50df459 79de83d d6fab25 bc9bce5 d6fab25 cba5a66 d6fab25 cba5a66 d6fab25 b2f6cec d6fab25 cba5a66 d6fab25 cba5a66 d6fab25 5396710 d6fab25 3c9f801 d6fab25 cba5a66 d6fab25 5396710 cba5a66 d6fab25 cba5a66 b2f6cec cba5a66 b2f6cec 3c9f801 dbe879e b2f6cec cba5a66 fd79e98 3123910 dbe879e 3123910 dbe879e 3eb35e9 dbe879e 1e10395 18d3583 05120d5 18d3583 05120d5 18d3583 05120d5 18d3583 05120d5 18d3583 05120d5 18d3583 05120d5 18d3583 05120d5 18d3583 dbe879e 1e10395 dbe879e 1e10395 dbe879e 1e10395 6efae71 1e10395 6efae71 1e10395 6efae71 1e10395 6efae71 1e10395 dbe879e 1e10395 dbe879e 1e10395 dbe879e 3eb35e9 dbe879e 1e10395 dbe879e 1e10395 dbe879e 3eb35e9 4a525e8 3eb35e9 1e10395 3eb35e9 1e10395 3eb35e9 1e10395 3eb35e9 1e10395 628312d 1e10395 3eb35e9 1e10395 628312d 3eb35e9 628312d 3eb35e9 dbe879e 0e4a811 dbe879e 61c709c fd44c51 61c709c fd44c51 61c709c fd44c51 61c709c dbe879e 1e10395 dbe879e 6efae71 1e10395 6efae71 dbe879e 3123910 959633e dbe879e 61c709c e4002ce 61c709c 370041e 61c709c e4002ce 959633e c082084 dbe879e 959633e dbe879e e8ece8a 959633e dbe879e 959633e c082084 dbe879e 959633e dbe879e 959633e d6fab25 5396710 cba5a66 d6fab25 52e8e7a d6fab25 b2f6cec d6fab25 cba5a66 d6fab25 030303e d6fab25 52e8e7a d6fab25 52e8e7a d6fab25 bc9bce5 d6fab25 cba5a66 d6fab25 852ee83 cba5a66 f8ccb74 0e4a811 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 fd44c51 f8ccb74 fd44c51 f8ccb74 e9b2d8c f8ccb74 e9b2d8c 8b74ef8 e9b2d8c dbe879e acfb205 f8ccb74 850db0e 61c709c f8ccb74 61c709c f8ccb74 e38d3be f8ccb74 e38d3be f8ccb74 eb59e59 f8ccb74 5c70b17 f8ccb74 5c70b17 f8ccb74 eb59e59 f8ccb74 850db0e f8ccb74 acfb205 fd44c51 61c709c f8ccb74 fd44c51 f8ccb74 fd44c51 f8ccb74 61c709c f8ccb74 61c709c f8ccb74 61c709c f8ccb74 e9b2d8c f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 6c168d6 f8ccb74 e9b2d8c f8ccb74 a8a6287 f8ccb74 fd44c51 f8ccb74 fd44c51 f8ccb74 a8a6287 fd44c51 a8a6287 fd44c51 a8a6287 fd44c51 a8a6287 f8ccb74 a8a6287 f8ccb74 a8a6287 f8ccb74 fd44c51 f8ccb74 e9b2d8c f8ccb74 1e10395 f8ccb74 1e10395 e9b2d8c 1e10395 e9b2d8c dbe879e e9b2d8c dbe879e e9b2d8c f8ccb74 e9b2d8c f8ccb74 e9b2d8c 05120d5 18d3583 05120d5 e9b2d8c 05120d5 1e10395 18d3583 1e10395 e9b2d8c dbe879e e9b2d8c f8ccb74 1e10395 18d3583 1e10395 18d3583 1e10395 18d3583 1e10395 18d3583 e9b2d8c f8ccb74 18d3583 f8ccb74 1e10395 f8ccb74 e9b2d8c f8ccb74 e9b2d8c 5a91767 cba5a66 5a91767 b2f6cec cba5a66 f20d300 cba5a66 f20d300 cba5a66 f20d300 cba5a66 f20d300 d6fab25 5a91767 c0dd18b d6fab25 5a91767 c0dd18b d6fab25 c0dd18b d6fab25 c0dd18b d6fab25 c0dd18b d6fab25 c0dd18b d6fab25 5a91767 c0dd18b 5a91767 c0dd18b 5a91767 d6fab25 c0dd18b 5a91767 c0dd18b 5a91767 c0dd18b d6fab25 df959f2 c0dd18b 5a91767 c0dd18b 5a91767 bbeb85a c0dd18b d547805 c0dd18b d6fab25 d547805 c0dd18b d547805 c0dd18b 5a91767 c0dd18b 5a91767 c0dd18b 61c709c 5a91767 c0dd18b 5a91767 61c709c c0dd18b 5a91767 46b58d7 c0dd18b 5a91767 c0dd18b 5a91767 c0dd18b f20d300 c0dd18b d547805 5a91767 d547805 d6fab25 d547805 4bcb970 d547805 4bcb970 d547805 4bcb970 d547805 c0dd18b d6fab25 5a91767 c0dd18b d6fab25 5a91767 46b58d7 5a91767 61c709c 3c9f801 | 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 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 | """
Khalifa University Library AI Agent
MCP-style tool-calling backend with RAG, PRIMO, PubMed, Google Scholar, Consensus
Tools:
- search_primo: Search KU Library catalog
- search_pubmed: Search biomedical literature
- search_scholar: Search Google Scholar
- search_consensus: Search Consensus (research papers)
- get_library_info: RAG from KU library knowledge base
Environment variables (HF Space Secrets):
OPENAI_API_KEY — required (embeddings + ChatGPT)
ANTHROPIC_API_KEY — optional (Claude answers)
PRIMO_API_KEY — required (PRIMO search)
"""
import os
import json
import re
import glob
import time
import sqlite3
import hashlib
import secrets
from datetime import datetime
from contextlib import asynccontextmanager
from fastapi import FastAPI, Request, Response, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import HTMLResponse, RedirectResponse
from pydantic import BaseModel
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain_community.vectorstores import FAISS
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.schema import Document
import httpx
# ===== CONFIG =====
KNOWLEDGE_DIR = "knowledge"
# Use /data for persistence across HF Space restarts — fall back to local if /data unavailable
FAISS_INDEX_PATH = "/data/faiss_index" if os.path.exists("/data") else "faiss_index"
DB_PATH = "/data/analytics.db" if os.path.exists("/data") else "analytics.db"
CHUNK_SIZE = 800
CHUNK_OVERLAP = 100
TOP_K = 7
# ===== MEDICAL TOPIC DETECTION =====
MEDICAL_KEYWORDS = [
'medicine','medical','health','clinical','nursing','pharmacy','biomedical',
'anatomy','physiology','epidemiology','oncology','cardiology','neuroscience',
'genomics','pathology','surgery','mental health','nutrition','public health',
'disease','therapy','treatment','diagnosis','patient','drug','pharmaceutical',
'hospital','cancer','diabetes','heart','brain','cell biology','immunology',
'biochemistry','microbiology','virology','pediatric','psychiatric','dental',
'ophthalmology','radiology','anesthesia','emergency medicine','geriatric'
]
CURRENT_INFO_RE = re.compile(r"\b(current|currently|latest|today|recent|recently|now|this week|this month|news|update|updated|president|prime minister|ceo|minister|king|ruler|who won)\b", re.IGNORECASE)
RESEARCH_CUE_RE = re.compile(r"\b(find|search|look for|show me|get me|give me|locate|recommend|suggest|articles?|papers?|books?|journals?|studies|literature|research|systematic review|evidence|sources|database|databases|peer reviewed|open access|advances?|approach(es)?|method(s|ology)?|technique[s]?|application[s]?|framework[s]?|model[s]?|algorithm[s]?|publication[s]?|recent work|current state|state of the art)\b", re.IGNORECASE)
LIBRARY_CUE_RE = re.compile(r"\b(library|librarian|borrow|loan|renew|fine|study room|room booking|reserve a room|account|my library|interlibrary|ill|khazna|orcid|open access|apc|refworks|libkey|hours|location|contact|visitor|alumni|database access|off campus|remote access)\b", re.IGNORECASE)
MEDICAL_SEARCH_RE = re.compile(r"\b(pubmed|embase|cinahl|clinicalkey|cochrane|uptodate|medline|systematic review|clinical trial|biomedical literature|medical literature)\b", re.IGNORECASE)
SOCIAL_RE = re.compile(r"^(hi|hello|hey|good morning|good afternoon|good evening|how are you|how old are you|who are you|what are you|are you a bot|are you a robot|what can you do|thanks|thank you|ok|okay|bye|goodbye|lol|haha|hehe|you are silly|are you silly|are you dumb|stupid bot|dumb bot|idiot bot|joke|tell me a joke)\b", re.IGNORECASE)
PURE_GREETING_RE = re.compile(r"^(hi|hello|hey|good morning|good afternoon|good evening)[!.\s]*$", re.IGNORECASE)
HOURS_RE = re.compile(
r"\b("
r"library hours?|opening hours?|closing hours?|open hours?|"
r"open today|open tomorrow|open now|open on|open this|"
r"closed today|closed tomorrow|closed on|closed this|"
r"when.*open|when.*close|when.*does.*library|"
r"what time.*open|what time.*close|what time.*library|"
r"opening time|closing time|"
r"is.*library.*open|is.*campus.*library|is.*habshan.*open|is.*san.*open|"
r"is.*library.*closed|will.*library.*open|will.*it.*open|"
r"does.*library.*open|library.*open.*today|library.*close.*today|"
r"habshan.*open|san.*open|main campus.*open|"
r"ramadan hours?|weekend hours?|friday hours?|saturday hours?|sunday hours?|"
r"library.*schedule|schedule.*library|"
r"what.*hours|till.*when|until.*when|"
r"what day.*open|which day.*open|what day.*closed|which day.*closed"
r")\b",
re.IGNORECASE
)
KU_GENERAL_RE = re.compile(r"\b(khalifa university|ku)\b.*\b(admission|admissions|program|programs|degree|degrees|college|colleges|school|schools|tuition|fees|scholarship|scholarships|hostel|housing|transport|ranking|rankings|president|vice president|chancellor|application|apply|registrar|academic calendar|semester dates|campus map)\b|\b(admission|admissions|program|programs|degree|degrees|college|colleges|school|schools|tuition|fees|scholarship|scholarships|hostel|housing|transport|ranking|rankings|president|vice president|chancellor|application|apply|registrar|academic calendar|semester dates|campus map)\b.*\b(khalifa university|ku)\b", re.IGNORECASE)
LIBRARY_HOURS_URL = "https://library.ku.ac.ae/hours"
KU_MAIN_URL = "https://www.ku.ac.ae/"
GREETING_FOLLOWUP_RE = re.compile(r"^(yes|yes please|yes check|sure|okay|ok|yeah|yep|please do|go ahead)$", re.IGNORECASE)
# Campus / location questions — always answered with hardcoded text, never from web
KU_CAMPUS_RE = re.compile(
r"\b("
r"where is the library|where is.*library|library.*location|location.*library|"
r"which building|what building|which floor|"
r"main campus library|san campus library|habshan library|sas al nakhl library|"
r"arzanah library|"
r"library.*address|address.*library|"
r"how to get to.*library|how.*reach.*library|how.*find.*library|"
r"library.*campus(es)?|campus(es)?.*library|"
r"how many.*librar|how many.*branch|library.*branch(es)?|branch(es)?.*library|"
r"which campus.*library|library.*which campus|"
r"library.*map|map.*library"
r")\b",
re.IGNORECASE
)
def _last_assistant_message(history):
if not history:
return ""
for m in reversed(history):
if m.get("role") == "assistant":
return m.get("content", "") or ""
return ""
def _is_greeting_menu_followup(question: str, history) -> bool:
q = (question or "").strip()
if not GREETING_FOLLOWUP_RE.match(q):
return False
last = _last_assistant_message(history).lower()
return (
"i’m <strong>libbee</strong>" in last
or "i'm <strong>libbee</strong>" in last
or "are you looking for one of these" in last
or "khalifa university library ai assistant" in last
)
def _greeting_menu_clarify_answer() -> str:
return (
"Sure — what would you like help with?<br><br>"
"You can choose one of these or type your question directly:<br>"
"• Find articles and books<br>"
"• Get full text for an article<br>"
"• Submit an Interlibrary Loan (ILL) request<br>"
"• Contact a librarian<br>"
"• Check library hours"
)
def _normalize_social_text(text: str) -> str:
t = (text or "").strip().lower()
t = re.sub(r"[^\w\s]", "", t)
t = re.sub(r"\s+", " ", t).strip()
return t
def _fixed_social_answer(question: str) -> str | None:
q = _normalize_social_text(question)
if re.fullmatch(r"(how are you|hows it going|how are you doing|how do you do)", q):
return (
"I’m doing well, thank you. I’m <strong>LibBee</strong>, the Khalifa University Library AI Assistant, "
"and I’m here to help with articles, books, databases, full text, and library services."
)
if re.fullmatch(r"(who are you|what are you|tell me about yourself)", q):
return (
"I’m <strong>LibBee</strong>, the Khalifa University Library AI Assistant. "
"I can help you find articles and books, search databases, get full text, "
"answer library service questions, and guide you to the right librarian when needed."
)
if re.fullmatch(r"(what can you do|what do you do|how can you help|what help can you provide)", q):
return (
"I can help you search for articles and books, suggest databases, find full text, "
"answer questions about library services, and connect you with the right Khalifa University Library staff member."
)
if re.fullmatch(r"(are you a bot|are you a robot|are you ai|are you an ai)", q):
return (
"Yes — I’m an AI assistant for Khalifa University Library. "
"My job is to help with library research, databases, articles, books, full text, and library services."
)
if re.fullmatch(r"(thank you|thanks|thx)", q):
return "You’re welcome. I’m happy to help."
if re.fullmatch(r"(bye|goodbye|see you)", q):
return "Goodbye. If you need help later, I’ll be here."
return None
STAFF_DIRECTORY = [
# ── Library Leadership ───────────────────────────────────────────
{
"full_name": "Dr. Abdulla Al Hefeiti",
"role": "Library Director / Assistant Provost, Libraries",
"details": "He can help with strategic and institutional matters, library leadership, and partnerships. Email: abdulla.alhefeiti@ku.ac.ae | Phone: +971 2 312 3331",
"tokens": ["abdulla", "hefeiti", "abdulla al hefeiti", "al hefeiti"],
},
# ── Specialist Librarians ────────────────────────────────────────
{
"full_name": "Nikesh Narayanan",
"role": "Research & Access Services Librarian",
"details": (
"He can help with research support, Khazna institutional repository, ORCID, "
"SciVal, Scopus, research impact, AI tools for research, bibliometrics, "
"scholarly communication, and LibGuides. "
"Email: nikesh.narayanan@ku.ac.ae | Phone: +971 2 312 3980"
),
"tokens": ["nikesh", "narayanan", "nikesh narayanan"],
},
{
"full_name": "Rani Anand",
"role": "E-Resources Librarian",
"details": (
"She can help with database access problems, e-resources troubleshooting, "
"remote access, vendor issues, patents, e-books, and bibliometrics. "
"Email: rani.anand@ku.ac.ae | Phone: +971 2 312 3935"
),
"tokens": ["rani", "anand", "rani anand"],
},
{
"full_name": "Jason Fetty",
"role": "Medical Librarian",
"details": (
"He can help with medical and health sciences research, systematic reviews, "
"PubMed, Embase, CINAHL, UpToDate, and clinical databases. "
"Email: jason.fetty@ku.ac.ae | Phone: +971 2 312 4722"
),
"tokens": ["jason", "fetty", "jason fetty"],
},
{
"full_name": "Walter Brian Hall",
"role": "Digital & Technology Services Librarian / Systems Librarian",
"details": (
"He can help with library website, systems, technology, digital infrastructure, "
"ORCID, STEAM, coding, Open Access publishing, APC funding, "
"general library services, and public services. "
"Email: walter.hall@ku.ac.ae | Open Access: openaccess@ku.ac.ae | "
"Phone: +971 2 312 3163 | Mobile: 052-905-1491"
),
"tokens": ["walter", "brian", "hall", "walter brian hall", "walter hall", "brian hall"],
},
# ── Managers ────────────────────────────────────────────────────
{
"full_name": "Alia Al-Harrasi",
"role": "Manager, Technical Services",
"details": (
"She can help with cataloguing, metadata, acquisitions processing, and technical services. "
"Email: alia.alharrasi@ku.ac.ae | Phone: +971 2 312 3180"
),
"tokens": ["alia", "harrasi", "al-harrasi", "alia al harrasi", "alia al-harrasi"],
},
{
"full_name": "Muna Ahmad Mohammad Al Blooshi",
"role": "Manager, Public Services",
"details": (
"She can help with circulation, general library services on Main Campus, and public services. "
"Email: muna.alblooshi@ku.ac.ae | Phone: +971 2 312 4212"
),
"tokens": ["muna", "blooshi", "al blooshi", "al-blooshi", "muna al blooshi"],
},
# ── Library Information Senior Specialists / Specialists ─────────
{
"full_name": "Suaad Al Jneibi",
"role": "Library Information Senior Specialist",
"details": (
"She can help with Interlibrary Loan (ILL) requests, document delivery, "
"and general library questions. "
"Email: suaad.aljneibi@ku.ac.ae | Phone: +971 2 312 4278 | "
"Location: SAN Habshan Building, GF Room 5017AWS01"
),
"tokens": [
"suaad", "jneibi", "al jneibi", "suaad al jneibi",
"suuad", "suaad jneibi",
],
},
{
"full_name": "Shamsa Saeed Salem Alhosani",
"role": "Library Information Senior Specialist",
"details": (
"She can help with library cataloguing and general library assistance. "
"Email: shamsa.alhosani@ku.ac.ae | Phone: +971 2 312 3156 | "
"Location: SAN Habshan Building, GF Room 5013WS04"
),
"tokens": [
"shamsa", "alhosani", "shamsa alhosani",
"shamsa saeed", "shamsa salem", "shamsa saeed salem alhosani",
],
},
{
"full_name": "William Nathaniel Hutchinson III",
"role": "Library Information Senior Specialist",
"details": (
"He can help with library cataloguing and general library assistance. "
"Email: william.hutchinsoniii@ku.ac.ae | Phone: +971 2 312 3966"
),
"tokens": [
"william", "hutchinson", "william hutchinson",
"william nathaniel", "hutchinson iii", "william hutchinson iii",
],
},
{
"full_name": "Shaikha Mohamed Saleem Almenhali",
"role": "Library Information Specialist",
"details": (
"She can help with general library assistance. "
"Email: shaikha.almenhali@ku.ac.ae | Phone: +971 2 312 5874"
),
"tokens": [
"shaikha", "almenhali", "shaikha almenhali",
"shaikha mohamed", "shaikha saleem", "shaikha mohamed saleem almenhali",
],
},
{
"full_name": "Wilfredo Arcilla",
"role": "Library Information Specialist",
"details": (
"He can help with general library assistance. "
"Email: wilfredo.arcilla@ku.ac.ae | Phone: +971 2 312 3323 | "
"Location: SAN Habshan Building, GF Room 5011WS05"
),
"tokens": [
"wilfredo", "arcilla", "wilfredo arcilla",
],
},
{
"full_name": "Zainab Al Ali",
"role": "Library Information Senior Specialist",
"details": (
"She can help with general library assistance. "
"Email: zainab.alali@ku.ac.ae | Phone: +971 2 312 4321 | "
"Location: SAN Habshan Building, GF Room 5017AWS02"
),
"tokens": [
"zainab", "al ali", "zainab al ali", "alali", "zainab alali",
],
},
{
"full_name": "Khawla Alhemeiri",
"role": "Library Information Senior Specialist",
"details": (
"She can help with general library assistance. "
"Email: khawla.alhemeiri@ku.ac.ae | Phone: +971 2 312 3959 | "
"Location: Main E Building, 1st Floor, Room E01035"
),
"tokens": [
"khawla", "alhemeiri", "khawla alhemeiri",
],
},
{
"full_name": "Meera K. Alnaqbi",
"role": "Library Information Specialist",
"details": (
"She can help with acquisitions, new title requests, collection development, "
"and general library assistance. "
"Email: meera.alnaqbi@ku.ac.ae | Phone: +971 2 312 3375"
),
"tokens": [
"meera", "alnaqbi", "meera alnaqbi", "meera k alnaqbi",
],
},
]
def _normalize_name_query(text: str):
return [t for t in re.sub(r"[^a-z0-9 ]+", " ", (text or "").lower()).split() if t]
def _dedupe_keep_order(items):
seen = set()
out = []
for item in items:
if item and item not in seen:
seen.add(item)
out.append(item)
return out
def _title_case_name(name: str) -> str:
return re.sub(r"\s+", " ", (name or "").strip()).title()
def _build_staff_tokens(full_name: str):
honorifics = {"dr", "mr", "mrs", "ms", "prof"}
raw_tokens = _normalize_name_query(full_name)
core_tokens = [t for t in raw_tokens if t not in honorifics]
token_lists = [raw_tokens, core_tokens]
variants = []
for toks in token_lists:
if not toks:
continue
variants.extend(toks)
variants.append(" ".join(toks))
for n in (2, 3, 4):
if len(toks) >= n:
for i in range(len(toks) - n + 1):
variants.append(" ".join(toks[i:i+n]))
return _dedupe_keep_order(variants)
def _parse_staff_directory_text(text: str):
staff_entries = []
if not text:
return staff_entries
lines = [line.rstrip() for line in text.splitlines()]
i = 0
while i < len(lines):
line = lines[i].strip()
is_name_line = (
line
and line == line.upper()
and not line.startswith("===")
and not line.startswith("SOURCE:")
and not line.startswith("TITLE:")
and any(ch.isalpha() for ch in line)
and len(line.split()) <= 10
)
if not is_name_line:
i += 1
continue
name_line = line
block = []
i += 1
while i < len(lines):
nxt = lines[i].strip()
next_is_name = (
nxt
and nxt == nxt.upper()
and not nxt.startswith("===")
and not nxt.startswith("SOURCE:")
and not nxt.startswith("TITLE:")
and any(ch.isalpha() for ch in nxt)
and len(nxt.split()) <= 10
)
if next_is_name:
break
block.append(nxt)
i += 1
role = ""
email = ""
phone = ""
mobile = ""
location = ""
best_for = ""
schedule = ""
extra_bits = []
for raw in block:
if not raw or raw.startswith("==="):
continue
low = raw.lower()
if raw.startswith("Title:"):
role = raw.split(":", 1)[1].strip()
elif raw.startswith("Email:"):
email = raw.split(":", 1)[1].strip()
elif raw.startswith("Phone:") or raw.startswith("Work Phone:"):
phone = raw.split(":", 1)[1].strip()
elif raw.startswith("Mobile:"):
mobile = raw.split(":", 1)[1].strip()
elif raw.startswith("Location:"):
location = raw.split(":", 1)[1].strip()
elif raw.startswith("Best for:"):
best_for = raw.split(":", 1)[1].strip()
elif raw.startswith("Schedule appointment:"):
schedule = raw.split(":", 1)[1].strip()
elif any(low.startswith(prefix) for prefix in ["linkedin:", "orcid:"]):
extra_bits.append(raw)
else:
extra_bits.append(raw)
details_parts = []
if best_for:
details_parts.append(f"Best for: {best_for}")
if email:
details_parts.append(f"Email: {email}")
if phone:
details_parts.append(f"Phone: {phone}")
if mobile:
details_parts.append(f"Mobile: {mobile}")
if location:
details_parts.append(f"Location: {location}")
if schedule:
details_parts.append(f"Schedule appointment: {schedule}")
details_parts.extend(extra_bits)
full_name = _title_case_name(name_line)
staff_entries.append({
"full_name": full_name,
"role": role or "Library staff member",
"details": " | ".join(_dedupe_keep_order(details_parts)),
"tokens": _build_staff_tokens(full_name),
"source_title": "Khalifa University Library Staff Directory and Contacts",
"source": "https://library.ku.ac.ae/librarystaff",
})
return staff_entries
def _load_staff_directory_from_kb():
entries = []
try:
for filepath in glob.glob(os.path.join(KNOWLEDGE_DIR, "*.txt")):
name = os.path.basename(filepath).lower()
if "staff" not in name and "contact" not in name:
continue
with open(filepath, "r", encoding="utf-8") as f:
content = f.read()
if "staff directory" not in content.lower() and "library staff" not in content.lower():
continue
entries.extend(_parse_staff_directory_text(content))
except Exception as e:
print(f"Staff KB parse error: {e}")
deduped = []
seen = set()
for entry in entries:
key = entry.get("full_name", "").lower()
if key and key not in seen:
seen.add(key)
deduped.append(entry)
return deduped
def _staff_lookup_candidates():
return kb_staff_directory or STAFF_DIRECTORY
def _role_key(name: str) -> str:
return re.sub(r"[^a-z0-9]+", " ", (name or "").lower()).strip()
def _build_role_aliases(staff: dict):
role = (staff.get("role") or "").strip()
details = (staff.get("details") or "").strip()
full_name = staff.get("full_name", "")
key = _role_key(full_name)
aliases = []
if role:
aliases.append(role)
role_l = role.lower()
parts = [p.strip() for p in re.split(r"[\/,]|\band\b", role_l) if p.strip()]
aliases.extend(parts)
if "manager" in role_l:
mgr_tail = re.sub(r"^manager\s*,?\s*", "", role_l).strip()
if mgr_tail:
aliases.extend([
mgr_tail,
f"{mgr_tail} librarian",
f"{mgr_tail.rstrip('s')} librarian",
f"{mgr_tail} manager",
])
if "librarian" in role_l:
base = role_l.replace("librarian", "").replace("/", " ").strip(" ,")
if base:
aliases.extend([
base,
f"{base} librarian",
f"{base.rstrip('s')} librarian",
])
best_for = ""
m = re.search(r"Best for:\s*([^|]+)", details, re.IGNORECASE)
if m:
best_for = m.group(1).strip()
if best_for:
aliases.extend([p.strip() for p in best_for.split(",") if p.strip()])
manual = {
"nikesh narayanan": [
"research librarian", "access services librarian", "research and access services librarian",
"research access librarian", "orcid librarian",
"research impact librarian", "bibliometrics librarian", "scholarly communication librarian",
"khazna librarian", "scival librarian",
],
"walter brian hall": [
"systems librarian", "system librarian", "library systems librarian",
"digital services librarian", "technology services librarian", "website librarian",
"technology librarian", "digital librarian",
"open access librarian", "open access", "oa librarian",
"apc librarian", "article processing charge",
"general librarian", "public services librarian",
],
"rani anand": [
"e resources librarian", "e-resources librarian", "electronic resources librarian",
"database librarian", "database access librarian", "resource access librarian",
],
"jason fetty": [
"medical librarian", "health sciences librarian", "clinical librarian",
"systematic review librarian", "pubmed librarian",
],
"alia al harrasi": [
"acquisitions librarian", "acquisition librarian", "technical services librarian",
"technical service librarian", "collection development librarian", "cataloguing librarian",
"metadata librarian",
],
"muna ahmad mohammad al blooshi": [
"public services librarian", "public service librarian", "circulation librarian",
"access services manager", "service desk librarian",
],
"dr abdulla al hefeiti": [
"library director", "director of library", "director libraries", "assistant provost libraries",
],
"abdulla al hefeiti": [
"library director", "director of library", "director libraries", "assistant provost libraries",
],
# ── New staff role aliases ──────────────────────────────────
"suaad al jneibi": [
"ill librarian", "interlibrary loan librarian", "ill specialist",
"ill contact", "document delivery", "ill staff",
],
"shamsa saeed salem alhosani": [
"cataloguing specialist", "cataloging specialist", "cataloguer",
"library cataloguing", "library cataloging",
],
"william nathaniel hutchinson iii": [
"cataloguing specialist", "cataloging specialist", "cataloguer",
"library cataloguing", "library cataloging",
],
"meera k alnaqbi": [
"acquisitions specialist", "acquisitions staff", "acquisition specialist",
"new title request", "book request", "purchase request",
"collection development specialist",
],
"meera k. alnaqbi": [
"acquisitions specialist", "acquisitions staff", "acquisition specialist",
"new title request", "book request", "purchase request",
"collection development specialist",
],
}
aliases.extend(manual.get(key, []))
normalized = []
for a in aliases:
a = re.sub(r"[^a-z0-9&/ +()-]+", " ", (a or "").lower())
a = re.sub(r"\s+", " ", a).strip(" ,")
if a:
normalized.append(a)
if a.endswith(" services"):
normalized.append(a[:-1])
if a.endswith(" service"):
normalized.append(a + " librarian")
if a.endswith(" resources"):
normalized.append(a[:-1])
return _dedupe_keep_order(normalized)
def _match_staff_role(question: str):
ql = re.sub(r"[^a-z0-9 ]+", " ", (question or "").lower())
ql = re.sub(r"\s+", " ", ql).strip()
if not ql:
return None
# Staff role lookup only makes sense for short questions asking WHO/CONTACT
# Long research questions (>12 words) should never trigger this
if len(ql.split()) > 12:
return None
# Also exclude questions that are clearly research searches, not staff lookups
research_intent_re = re.compile(
r"\b(i need|find me|looking for|search for|show me|give me|"
r"articles?|papers?|books?|literature|how effective|how does|"
r"what are|what is the|impact of|effect of|based on|published)\b"
)
if research_intent_re.search(ql):
return None
role_trigger_re = re.compile(
r"\b(librarian|library director|director|manager|services?|service|systems?|technology|website|"
r"research|access|open access|orcid|bibliometric|bibliometrics|public services?|"
r"technical services?|acquisitions?|collection development|catalogu(?:e|ing)|metadata|"
r"database|e resources|e-resources|electronic resources|medical|clinical|systematic review|"
r"circulation|interlibrary loan|ill|document delivery|"
r"cataloging|cataloguer|general library|general service)\b"
)
if not role_trigger_re.search(ql):
return None
best = None
best_score = 0
for staff in _staff_lookup_candidates():
score = 0
for alias in _build_role_aliases(staff):
if not alias:
continue
alias_words = alias.split()
if alias in ql:
score = max(score, 100 + len(alias_words))
elif len(alias_words) >= 2 and all(w in ql.split() for w in alias_words):
score = max(score, 70 + len(alias_words))
else:
overlap = sum(1 for w in alias_words if len(w) > 2 and w in ql.split())
if overlap >= 2:
score = max(score, 40 + overlap)
role_words = _normalize_name_query(staff.get("role", ""))
if role_words:
overlap = sum(1 for w in role_words if len(w) > 2 and w in ql.split())
if overlap >= 2:
score = max(score, 20 + overlap)
if score > best_score:
best_score = score
best = staff
return best if best_score >= 42 else None
def _match_staff_name(question: str):
tokens = _normalize_name_query(question)
if not tokens or len(tokens) > 5:
return None
ql = (question or "").strip().lower()
blocked = ["who is", "who handles", "who can help", "contact", "librarian", "library"]
if any(b in ql for b in blocked):
return None
for staff in _staff_lookup_candidates():
staff_tokens = set()
for tok in staff.get("tokens", []):
staff_tokens.update(_normalize_name_query(tok))
staff_tokens.add(" ".join(_normalize_name_query(tok)))
if all(tok in staff_tokens for tok in tokens):
return staff
return None
def _staff_name_answer(staff: dict, partial: str) -> str:
return (
f"Did you mean <strong>{staff['full_name']}</strong>?<br><br>"
f"<strong>{staff['full_name']}</strong> is the <strong>{staff['role']}</strong>. {staff['details']}"
)
def _staff_role_answer(staff: dict, question: str) -> str:
return (
f"For <strong>{question}</strong>, the best match is <strong>{staff['full_name']}</strong> — "
f"<strong>{staff['role']}</strong>.<br><br>{staff['details']}"
)
GROUNDED_LIBRARY_MAP = {
"ill": "interlibrary loan ILL document delivery request article request book unavailable not in collection borrow from another library Suaad Al Jneibi",
"fulltext": "full text libkey nomad article access pdf article not available no full text get full text request article",
"primo": "PRIMO discovery catalog holdings publication finder library has this journal books articles search catalog",
"circulation": "borrowing circulation renew renewal loan period due date hold request fines fees my library account checked out overdue",
"research_help": "Ask a Librarian research consultation reference help subject guides research skills instruction tours workshops",
"orcid_oa": "ORCID open access APC publishing research impact Scopus SciVal bibliometrics Walter Brian Hall openaccess@ku.ac.ae",
"database_access": "database access e-resources remote access off campus login vendor issue Rani Anand e-books patents",
"medical_help": "medical librarian Jason Fetty PubMed Embase CINAHL Cochrane UpToDate systematic review clinical databases",
"systems_help": "Walter Brian Hall systems website technology digital services library systems coding STEAM open access APC",
"acquisitions": "Alia Al-Harrasi Meera Alnaqbi acquisitions request title suggest a book collection development technical services cataloguing metadata",
"cataloguing": "Shamsa Alhosani William Hutchinson Alia Al-Harrasi cataloguing cataloging metadata technical services",
"rooms": "study room reserve a room room booking group study rooms libcal book room room reservation",
"visitors_alumni": "visitors visitor registration alumni alumni services external visitors access policy",
"printing": "printing scanning photocopying printers scanner copy services",
"refworks": "RefWorks citation manager reference management citations bibliography write n cite",
"hours_locations": "library hours opening hours closing hours locations main campus Habshan SAN Sas Al Nakhl",
"standards": "ASTM Compass IEEE standards ASME ASCE engineering standards",
"theses": "ProQuest Dissertations theses Khazna institutional repository dissertations theses",
"metrics": "impact factor journal citation reports JCR CiteScore Scopus SciVal metrics",
"chemistry": "ACS SciFindern RSC Reaxys chemistry database",
"physics": "APS AIP IOPScience physics database",
"engineering": "IEEE Xplore ACM ASCE ASME Knovel engineering databases",
}
# ===== GLOBALS =====
vectorstore = None
http_client = None
kb_staff_directory = []
# ===== ANALYTICS DB =====
def init_db():
conn = sqlite3.connect(DB_PATH)
conn.execute('''CREATE TABLE IF NOT EXISTS queries (
id INTEGER PRIMARY KEY AUTOINCREMENT,
timestamp TEXT, question TEXT, tool_used TEXT,
model TEXT, response_time REAL, result_count INTEGER,
error TEXT
)''')
conn.execute('''CREATE TABLE IF NOT EXISTS bot_config (
key TEXT PRIMARY KEY,
value TEXT,
updated_at TEXT
)''')
conn.execute('''CREATE TABLE IF NOT EXISTS feedback (
id INTEGER PRIMARY KEY AUTOINCREMENT,
timestamp TEXT, question TEXT, feedback TEXT, answer_excerpt TEXT, intent TEXT
)''')
# Set defaults if not exist
defaults = {
"welcome_message": "Hi! I'm the Khalifa University Library AI Assistant. I can help you find articles, books, databases, and answer questions about library services. How can I help you today?",
"bot_personality": "You are a helpful, friendly, and knowledgeable library assistant at Khalifa University, Abu Dhabi, UAE. KU = Khalifa University, NOT Kuwait University. Be concise (2-4 sentences). Always include relevant URLs.",
"custom_instructions": "",
"announcement": "",
"max_results": "5",
"default_model": "gpt",
"maintenance_mode": "false",
"maintenance_message": "The library chatbot is currently undergoing maintenance. Please try again later.",
}
for k, v in defaults.items():
conn.execute(
"INSERT OR IGNORE INTO bot_config (key, value, updated_at) VALUES (?, ?, ?)",
(k, v, datetime.utcnow().isoformat())
)
conn.commit()
conn.close()
def get_config(key, default=""):
try:
conn = sqlite3.connect(DB_PATH)
row = conn.execute("SELECT value FROM bot_config WHERE key=?", (key,)).fetchone()
conn.close()
return row[0] if row else default
except:
return default
def set_config(key, value):
conn = sqlite3.connect(DB_PATH)
conn.execute(
"INSERT OR REPLACE INTO bot_config (key, value, updated_at) VALUES (?, ?, ?)",
(key, value, datetime.utcnow().isoformat())
)
conn.commit()
conn.close()
# ===== ADMIN AUTH =====
# ADMIN_PASSWORD must be set as HF Space Secret — no insecure fallback
ADMIN_PASSWORD = os.environ.get("ADMIN_PASSWORD", "")
if not ADMIN_PASSWORD:
import warnings
warnings.warn("ADMIN_PASSWORD secret is not set. Admin dashboard will be inaccessible until configured.")
admin_sessions = {} # token -> expiry timestamp
START_TIME = time.time() # for uptime tracking
def create_session():
token = secrets.token_hex(32)
admin_sessions[token] = time.time() + 86400 # 24 hour session
return token
def verify_session(request: Request):
token = request.cookies.get("admin_session")
if not token or token not in admin_sessions:
return False
if time.time() > admin_sessions[token]:
del admin_sessions[token]
return False
return True
def log_query(question, tool, model, response_time, result_count=0, error=None):
try:
conn = sqlite3.connect(DB_PATH)
conn.execute(
'INSERT INTO queries (timestamp,question,tool_used,model,response_time,result_count,error) VALUES (?,?,?,?,?,?,?)',
(datetime.utcnow().isoformat(), question, tool, model, response_time, result_count, error)
)
conn.commit()
conn.close()
except Exception as e:
print(f"Log error: {e}")
# ===== RAG SETUP =====
def load_documents():
global kb_staff_directory
docs = []
files = glob.glob(os.path.join(KNOWLEDGE_DIR, "*.txt"))
kb_staff_directory = _load_staff_directory_from_kb()
if kb_staff_directory:
print(f"Loaded {len(kb_staff_directory)} staff entries from KB")
for filepath in files:
try:
with open(filepath, "r", encoding="utf-8") as f:
content = f.read()
lines = content.split("\n", 3)
source, title, text = "", "", content
for line in lines[:2]:
if line.startswith("SOURCE:"): source = line.replace("SOURCE:", "").strip()
elif line.startswith("TITLE:"): title = line.replace("TITLE:", "").strip()
if source or title: text = "\n".join(lines[2:]).strip()
docs.append(Document(page_content=text, metadata={"source": source, "title": title}))
except Exception as e:
print(f"Error loading {filepath}: {e}")
print(f"Loaded {len(docs)} documents")
return docs
def build_vectorstore(docs, force_rebuild=False):
embeddings = OpenAIEmbeddings(model="text-embedding-3-small")
index_file = os.path.join(FAISS_INDEX_PATH, "index.faiss")
# On startup: load existing index if available (saves cost + time)
if not force_rebuild and os.path.exists(index_file):
print(f"Loading existing FAISS index from {FAISS_INDEX_PATH}")
store = FAISS.load_local(FAISS_INDEX_PATH, embeddings, allow_dangerous_deserialization=True)
print(f"Loaded existing index with {store.index.ntotal} chunks")
return store
# Force rebuild: always re-embed all documents from scratch
print(f"Building new FAISS index from {len(docs)} documents...")
splitter = RecursiveCharacterTextSplitter(chunk_size=CHUNK_SIZE, chunk_overlap=CHUNK_OVERLAP)
chunks = splitter.split_documents(docs)
print(f"Split into {len(chunks)} chunks, embedding now...")
store = FAISS.from_documents(chunks, embeddings)
os.makedirs(FAISS_INDEX_PATH, exist_ok=True)
store.save_local(FAISS_INDEX_PATH)
print(f"FAISS index saved with {store.index.ntotal} chunks")
return store
# ===== TOOL: SEARCH PRIMO =====
def _boolean_to_primo_params(boolean_query: str) -> str:
"""
Convert Boolean string to PRIMO REST API multi-query format.
PRIMO API uses q= (not query=) and requires %20 encoded spaces.
("AI" OR "ML") AND ("cancer") AND ("deep learning")
→ q=any,contains,AI%20OR%20ML,AND&q=any,contains,cancer,AND&q=any,contains,deep%20learning
Single concept: q=any,contains,machine%20learning
"""
from urllib.parse import quote
# If no Boolean operators — simple single query
if not (re.search(r'\bAND\b', boolean_query) and '(' in boolean_query):
clean = boolean_query.replace('"', '').strip()
return f"q=any,contains,{quote(clean)}"
# Split on top-level AND
groups = []
depth = 0
chunk = ""
for char in boolean_query:
if char == '(':
depth += 1
chunk += char
elif char == ')':
depth -= 1
chunk += char
else:
chunk += char
if depth == 0 and chunk.upper().endswith(' AND '):
groups.append(chunk[:-5].strip())
chunk = ""
if chunk.strip():
groups.append(chunk.strip())
if not groups:
clean = boolean_query.replace('"', '').strip()
return f"q=any,contains,{quote(clean)}"
primo_params = []
for i, group in enumerate(groups):
g = group.strip()
# Remove outer parentheses
if g.startswith('(') and g.endswith(')'):
g = g[1:-1].strip()
# Remove double quotes
g = g.replace('"', '')
# Normalise OR spacing
g = re.sub(r'\s+OR\s+', ' OR ', g).strip()
if not g:
continue
# URL encode properly — spaces become %20
encoded = quote(g)
suffix = ',AND' if i < len(groups) - 1 else ''
primo_params.append(f"q=any,contains,{encoded}{suffix}")
return '&'.join(primo_params) if primo_params else f"q=any,contains,{quote(boolean_query.replace(chr(34), '').strip())}"
async def tool_search_primo(query, limit=5, peer_reviewed=False, open_access=False, year_from=None, year_to=None):
"""
query can be either:
- a Boolean string: ("AI" OR "ML") AND ("cancer") — converted to multi-query PRIMO format
- a plain keyword string: "machine learning cancer" — sent as single query
"""
api_key = os.environ.get("PRIMO_API_KEY")
if not api_key: return {"error": "PRIMO_API_KEY not configured", "results": []}
# Build PRIMO query params — multi-query format preserves all concepts
if re.search(r'\b(AND|OR)\b', query) and '(' in query:
# Boolean string — convert to multi-query format
query_params = _boolean_to_primo_params(query)
else:
# Plain keywords — single query, URL encoded
from urllib.parse import quote
query_params = f"query=any,contains,{quote(query.strip())}"
vid = "971KUOSTAR_INST:KU"
facets = ""
if peer_reviewed: facets += "&qInclude=facet_tlevel,exact,peer_reviewed"
if open_access: facets += "&qInclude=facet_tlevel,exact,open_access"
if year_from or year_to:
yf = year_from or "1900"
yt = year_to or str(datetime.now().year)
facets += f"&multiFacets=facet_searchcreationdate,include,{yf}%7C,%7C{yt}"
base = "https://api-eu.hosted.exlibrisgroup.com/primo/v1/search"
qs = f"?vid={vid}&tab=Everything&scope=MyInst_and_CI&{query_params}&lang=en&sort=rank&limit={limit}&offset=0&mode=advanced&apikey={api_key}{facets}"
async with httpx.AsyncClient(timeout=15) as client:
for region in ["api-eu", "api-na", "api-ap"]:
url = base.replace("api-eu", region) + qs
try:
r = await client.get(url, headers={"Accept": "application/json"})
if r.status_code == 200:
data = r.json()
total = data.get("info", {}).get("total", 0)
results = []
for doc in data.get("docs", []):
pnx = doc.get("pnx", {})
d = pnx.get("display", {})
a = pnx.get("addata", {})
s = pnx.get("search", {})
c = pnx.get("control", {})
l = pnx.get("links", {})
record_id = (c.get("recordid") or [None])[0]
title = (d.get("title") or ["Untitled"])[0]
primo_url = _build_primo_search_link(title, record_id)
results.append({
"record_id": record_id,
"title": title,
"creator": "; ".join(d.get("creator") or d.get("contributor") or []) or "Unknown",
"date": (s.get("creationdate") or a.get("risdate") or a.get("date") or [""])[0],
"type": (d.get("type") or [""])[0],
"source": (d.get("source") or a.get("jtitle") or [""])[0],
"description": ((d.get("description") or [""])[0] or "")[:400],
"doi": (a.get("doi") or [None])[0],
"primo_url": primo_url,
"link": ((l.get("openurl") or l.get("linktorsrc") or [None])[0]) or primo_url,
"open_access": (d.get("oa") or [""])[0] == "free_for_read",
"_source": "PRIMO",
})
return {"total": total, "results": results, "source": "PRIMO"}
except Exception:
continue
return {"error": "PRIMO API unavailable", "results": [], "source": "PRIMO"}
# ===== MCP HELPER =====
async def _call_mcp(api_key, prompt, mcp_url, mcp_name, timeout=30):
"""Call Anthropic API with an MCP server. Returns text content or raises."""
async with httpx.AsyncClient(timeout=timeout) as client:
r = await client.post(
"https://api.anthropic.com/v1/messages",
headers={
"Content-Type": "application/json",
"x-api-key": api_key,
"anthropic-version": "2023-06-01",
"anthropic-beta": "mcp-client-2025-04-04",
},
json={
"model": "claude-sonnet-4-20250514",
"max_tokens": 2000,
"messages": [{"role": "user", "content": prompt}],
"mcp_servers": [{"type": "url", "url": mcp_url, "name": mcp_name}],
},
)
if r.status_code != 200:
raise Exception(f"Anthropic API: {r.status_code}")
data = r.json()
parts = []
for block in data.get("content", []):
if block.get("type") == "text":
parts.append(block["text"])
elif block.get("type") == "mcp_tool_result":
for item in block.get("content", []):
if item.get("type") == "text":
parts.append(item["text"])
return "\n".join(parts)
async def _parse_results(api_key, raw_text, source_name, limit=5):
"""Parse raw MCP results into structured JSON."""
async with httpx.AsyncClient(timeout=20) as client:
r = await client.post(
"https://api.anthropic.com/v1/messages",
headers={"Content-Type": "application/json", "x-api-key": api_key, "anthropic-version": "2023-06-01"},
json={
"model": "claude-sonnet-4-20250514", "max_tokens": 2000,
"messages": [{"role": "user", "content": f"""Extract search results into a JSON array. Each item: title, creator, date, source, doi, link, description, type. Return ONLY valid JSON array. No markdown. If none, return [].
Text:
{raw_text[:3000]}"""}],
},
)
if r.status_code != 200: return []
text = "".join(b["text"] for b in r.json().get("content", []) if b.get("type") == "text").strip()
if text.startswith("```"): text = text.split("\n", 1)[1].rsplit("```", 1)[0].strip()
results = json.loads(text)
for item in results:
item["_source"] = source_name
if item.get("doi") and not item.get("link"): item["link"] = f"https://doi.org/{item['doi']}"
return results[:limit]
# ===== TOOL: SEARCH PUBMED (direct NCBI E-utilities — free, reliable) =====
async def tool_search_pubmed(query, limit=5):
base = "https://eutils.ncbi.nlm.nih.gov/entrez/eutils"
try:
async with httpx.AsyncClient(timeout=15) as client:
r = await client.get(f"{base}/esearch.fcgi?db=pubmed&term={query}&retmax={limit}&retmode=json&sort=relevance")
if r.status_code != 200: return {"error": "PubMed search failed", "results": [], "source": "PubMed"}
ids = r.json().get("esearchresult", {}).get("idlist", [])
if not ids: return {"total": 0, "results": [], "source": "PubMed"}
r2 = await client.get(f"{base}/esummary.fcgi?db=pubmed&id={','.join(ids)}&retmode=json")
if r2.status_code != 200: return {"error": "PubMed fetch failed", "results": [], "source": "PubMed"}
data = r2.json().get("result", {})
results = []
for pid in ids:
rec = data.get(pid, {})
if not isinstance(rec, dict): continue
authors = ", ".join(a.get("name", "") for a in rec.get("authors", [])[:3])
if len(rec.get("authors", [])) > 3: authors += " et al."
results.append({
"title": rec.get("title", ""), "creator": authors, "date": rec.get("pubdate", ""),
"source": rec.get("fulljournalname", rec.get("source", "")),
"doi": rec.get("elocationid", "").replace("doi: ", "") if "doi:" in rec.get("elocationid", "") else None,
"pmid": pid, "link": f"https://pubmed.ncbi.nlm.nih.gov/{pid}/",
"type": "Journal Article", "_source": "PubMed",
})
total = int(r.json().get("esearchresult", {}).get("count", 0))
return {"total": total, "results": results, "source": "PubMed"}
except Exception as e:
return {"error": f"PubMed: {str(e)}", "results": [], "source": "PubMed"}
# ===== TOOL: SEARCH CONSENSUS (via Semantic Scholar with consensus framing) =====
async def tool_search_consensus(query, limit=5):
"""
Consensus.app requires OAuth so we can't call it directly.
Instead we search Semantic Scholar with the same query and return
results alongside a direct Consensus deep-link so the user can
also check the AI-synthesized answer there.
"""
scholar = await tool_search_scholar(query, limit)
scholar["source"] = "Consensus / Semantic Scholar"
scholar["consensus_url"] = f"https://consensus.app/results/?q={query}"
scholar["message"] = "Results from Semantic Scholar. For AI-synthesized consensus, click the Consensus link."
return scholar
# ===== TOOL: SEARCH SEMANTIC SCHOLAR (direct API — free, no auth) =====
async def tool_search_scholar(query, limit=5):
"""Search 200M+ papers via Semantic Scholar API. Free, no API key needed."""
url = "https://api.semanticscholar.org/graph/v1/paper/search/bulk"
params = {
"query": query,
"limit": min(limit, 20),
"fields": "title,authors,year,venue,externalIds,abstract,citationCount,openAccessPdf,publicationTypes",
}
try:
async with httpx.AsyncClient(timeout=15) as client:
r = await client.get(url, params=params, headers={"Accept": "application/json"})
if r.status_code != 200:
return {"error": f"Semantic Scholar API: {r.status_code}", "results": [], "source": "Semantic Scholar"}
data = r.json()
total = data.get("total", 0)
results = []
for paper in data.get("data", [])[:limit]:
authors = ", ".join(a.get("name", "") for a in (paper.get("authors") or [])[:3])
if len(paper.get("authors") or []) > 3:
authors += " et al."
ext_ids = paper.get("externalIds") or {}
doi = ext_ids.get("DOI")
link = None
if paper.get("openAccessPdf", {}).get("url"):
link = paper["openAccessPdf"]["url"]
elif doi:
link = f"https://doi.org/{doi}"
else:
s2_id = paper.get("paperId")
if s2_id:
link = f"https://www.semanticscholar.org/paper/{s2_id}"
results.append({
"title": paper.get("title", ""),
"creator": authors,
"date": str(paper.get("year", "")),
"source": paper.get("venue", ""),
"description": ((paper.get("abstract") or "")[:400]),
"doi": doi,
"link": link,
"citations": paper.get("citationCount", 0),
"type": (paper.get("publicationTypes") or ["Article"])[0] if paper.get("publicationTypes") else "Article",
"open_access": bool(paper.get("openAccessPdf")),
"_source": "Semantic Scholar",
})
return {"total": total, "results": results, "source": "Semantic Scholar"}
except Exception as e:
return {"error": f"Semantic Scholar: {str(e)}", "results": [], "source": "Semantic Scholar"}
async def tool_library_info(question, history=None, model="gpt"):
if not vectorstore:
return {"answer": "Knowledge base not initialized.", "sources": [], "has_answer": False}
# Build search query — use question + history context
base_query = question
if history:
history_text = "\n".join(f"{'User' if m['role']=='user' else 'Assistant'}: {m['content']}" for m in history[-3:])
base_query = f"{history_text}\n{question}"
# ── Semantic query expansion via lightweight LLM interpreter + grounded institutional map ──
interp = await _interpret_semantics(question, history or [])
grounding_text = " ".join(GROUNDED_LIBRARY_MAP.get(k, "") for k in interp.get("grounding_keys", []))
canonical_text = " ".join(interp.get("canonical_terms", []))
expanded_query = " ".join(x for x in [base_query, canonical_text, grounding_text] if x).strip()
# ── FAISS scored search ──
docs_with_scores = vectorstore.similarity_search_with_score(expanded_query, k=TOP_K)
if not docs_with_scores:
return {"answer": "", "sources": [], "has_answer": False}
best_score = min(score for _, score in docs_with_scores)
print(f"RAG best_score={best_score:.3f} for: {question[:60]}")
# More forgiving threshold for library-service questions; still reject clearly unrelated queries
score_threshold = 1.45
if interp.get("intent_hint") == "library_info":
score_threshold = 1.7
if best_score > score_threshold:
print(f"RAG skipped — score {best_score:.3f} exceeds {score_threshold:.2f} threshold")
return {"answer": "", "sources": [], "has_answer": False}
docs = [doc for doc, _ in docs_with_scores]
context = "\n\n".join(d.page_content for d in docs)
# If score is moderate (0.8–1.4), the match may be close but not exact.
# Instruct the LLM to add a "Did you mean...?" clarification if it interprets the question loosely.
moderate_match = 0.8 < best_score <= 1.4
did_you_mean_instruction = """
- If you are answering a question that uses different wording from what's in the context
(e.g. user asked "research librarian" but context says "Research and Access Services Librarian"),
start your answer with: "Did you mean [exact title from context]? If so, here is the information:"
then give the answer.""" if moderate_match else ""
prompt = f"""You are the Khalifa University Library AI Assistant in Abu Dhabi, UAE.
KU means Khalifa University, NOT Kuwait University.
RULES — follow exactly:
1. Answer ONLY using the context provided below.
2. If the context has relevant information → answer in 2-4 sentences, include URLs if present.{did_you_mean_instruction}
3. If the context does NOT contain the answer → output the single word: NO_LIBRARY_ANSWER
- Do NOT write "I'm sorry"
- Do NOT write "I don't have information"
- Do NOT suggest contacting the library
- Do NOT apologise
- ONLY output: NO_LIBRARY_ANSWER
Context:
{context}
Question: {question}
Answer:"""
# Respect the model selection — use Claude if requested, GPT otherwise
use_claude = model == "claude" and os.environ.get("ANTHROPIC_API_KEY")
if use_claude:
from langchain_anthropic import ChatAnthropic
llm = ChatAnthropic(model="claude-haiku-4-5-20251001", temperature=0.2, max_tokens=500)
else:
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.2, max_tokens=500)
response = llm.invoke(prompt)
answer = response.content.strip()
# LLM signalled it couldn't answer from context
if answer == "NO_LIBRARY_ANSWER" or answer.startswith("NO_LIBRARY_ANSWER"):
return {"answer": "", "sources": [], "has_answer": False}
# Catch cases where LLM ignored the NO_LIBRARY_ANSWER instruction
# and generated a polite "I don't have info" response instead
# Normalise smart/curly apostrophes to straight before pattern matching
answer_normalised = answer.lower().replace("’", "'").replace("‘", "'").replace("“", '"').replace("”", '"')
no_info_patterns = [
"i don't have",
"i do not have",
"i'm sorry",
"i am sorry",
"i apologize",
"i apologise",
"no specific information",
"not available in",
"not in the context",
"context does not contain",
"context doesn't contain",
"i cannot find",
"i could not find",
"don't have information",
"do not have information",
"unable to find",
"unable to provide",
"not found in",
"no information",
"for further inquiries",
"for more information, please contact",
"please visit ask",
"please contact the library",
"recommend checking",
"i'd suggest",
"i would suggest contacting",
]
if any(p in answer_normalised for p in no_info_patterns):
print(f"RAG no-info pattern matched: '{answer_normalised[:80]}'")
return {"answer": "", "sources": [], "has_answer": False}
sources = []
seen = set()
for d in docs:
src = d.metadata.get("source", "")
title = d.metadata.get("title", "")
key = src or title
if key and key not in seen:
seen.add(key)
sources.append({"title": title, "source": src})
return {"answer": answer, "sources": sources, "has_answer": True}
# ===== DETECT MEDICAL TOPIC =====
def is_medical_topic(text):
lower = text.lower()
return any(kw in lower for kw in MEDICAL_KEYWORDS)
# ===== STARTUP =====
@asynccontextmanager
async def lifespan(app: FastAPI):
global vectorstore, http_client
print("=== Starting KU Library AI Agent ===")
init_db()
docs = load_documents()
if docs:
vectorstore = build_vectorstore(docs)
print(f"Vector store ready with {vectorstore.index.ntotal} vectors")
http_client = httpx.AsyncClient(timeout=15)
yield
await http_client.aclose()
print("Shutting down...")
# ===== FASTAPI =====
app = FastAPI(title="KU Library AI Agent", lifespan=lifespan)
app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
# ===== MODELS =====
class ChatMessage(BaseModel):
role: str
content: str
class SearchRequest(BaseModel):
query: str
source: str = "primo" # primo, pubmed, scholar, consensus, all
model: str = "gpt"
limit: int = 5
peer_reviewed: bool = False
open_access: bool = False
year_from: str | None = None
year_to: str | None = None
class RAGRequest(BaseModel):
question: str
model: str = "gpt"
history: list[ChatMessage] = []
class AgentRequest(BaseModel):
question: str
model: str = "gpt"
history: list[ChatMessage] = []
# ===== ENDPOINTS =====
@app.get("/")
def health():
return {
"status": "ok",
"vectorstore_ready": vectorstore is not None,
"tools": ["search_primo", "search_pubmed", "search_scholar", "search_consensus", "get_library_info"],
"endpoints": ["/rag", "/search", "/agent", "/general", "/config", "/year"],
"models": {
"gpt": bool(os.environ.get("OPENAI_API_KEY")),
"claude": bool(os.environ.get("ANTHROPIC_API_KEY")),
},
}
# Fix 6: authoritative server-side year — frontend uses this instead of client device clock
@app.get("/year")
def get_year():
now = datetime.utcnow()
return {
"year": now.year,
"month": now.month,
"date": now.strftime("%Y-%m-%d"),
}
# ---- Spell correction + Boolean building (uses HF Space OpenAI — separate quota from Cloudflare) ----
class CorrectRequest(BaseModel):
query: str
year: int = 2026
def _extract_json_object(text: str) -> dict:
s, e = text.find("{"), text.rfind("}")
if s == -1 or e == -1 or e <= s:
raise ValueError("No JSON object found")
return json.loads(text[s:e+1])
def _make_boolean(text: str) -> str:
"""Build a basic Boolean from text by extracting meaningful phrase groups."""
import re as re2
stop = re2.compile(
r'^(of|on|in|the|a|an|and|or|for|to|with|by|from|at|is|are|was|were|be|been|'
r'being|have|has|had|do|does|did|will|would|could|should|may|its|this|that|'
r'these|those|about|impact|effect|role|influence|importance|use|application|'
r'study|analysis|review|what|how|why|when|where|which)$', re2.IGNORECASE
)
words = text.split()
concepts, phrase = [], []
for w in words:
clean = w.strip("'\"")
if not stop.match(clean) and len(clean) > 2:
phrase.append(clean)
else:
if phrase:
concepts.append(' '.join(phrase))
phrase = []
if phrase:
concepts.append(' '.join(phrase))
if len(concepts) >= 2:
return ' AND '.join(f'"{c}"' for c in concepts)
if len(concepts) == 1:
return f'"{concepts[0]}"'
return f'"{text}"'
def _clean_database_keywords(boolean_query: str) -> str:
return re.sub(r'\s+', ' ', re.sub(r'\b(AND|OR|NOT)\b|[()"]', ' ', boolean_query, flags=re.IGNORECASE)).strip()
def _light_strip_retrieval_boilerplate(text: str) -> str:
cleaned = re.sub(r'^\s*(please\s+)?(?:can you|could you|would you)\s+', '', (text or '').strip(), flags=re.IGNORECASE)
cleaned = re.sub(r'^\s*(please\s+)?help me\s+', '', cleaned, flags=re.IGNORECASE)
cleaned = re.sub(r'^\s*please\s+', '', cleaned, flags=re.IGNORECASE)
cleaned = re.sub(r'\s+(please|thanks|thank you|asap)$', '', cleaned, flags=re.IGNORECASE)
return re.sub(r'\s+', ' ', cleaned).strip()
def _build_primo_search_link(title: str, record_id: str | None = None) -> str | None:
base = 'https://khalifa.primo.exlibrisgroup.com/discovery'
if record_id:
return f"{base}/fulldisplay?docid={quote(str(record_id))}&vid=971KUOSTAR_INST:KU"
clean_title = (title or '').strip()
if not clean_title:
return None
return f"{base}/search?query=any,contains,{quote(clean_title)}&tab=Everything&search_scope=MyInst_and_CI&vid=971KUOSTAR_INST:KU&lang=en"
def _parse_year_value(value) -> int | None:
m = re.search(r'(19|20)\d{2}', str(value or ''))
return int(m.group(0)) if m else None
def _is_article_like(result: dict) -> bool:
type_text = str(result.get('type') or '').lower()
source_text = str(result.get('source') or '').lower()
if any(k in type_text for k in ['article', 'journal', 'review', 'paper', 'study']):
return True
return bool(source_text)
def _normalize_result_links(result: dict) -> dict:
result = dict(result or {})
record_id = result.get('record_id') or result.get('id')
title = result.get('title') or ''
primo_url = result.get('primo_url') or _build_primo_search_link(title, record_id if result.get('_source') == 'PRIMO' else None)
if primo_url:
result['primo_url'] = primo_url
if result.get('_source') == 'PRIMO' and not result.get('link'):
result['link'] = primo_url
return result
def _choose_verified_article(results: list[dict], year_from=None, year_to=None) -> dict | None:
candidates = []
y_from = int(year_from) if str(year_from or '').isdigit() else None
y_to = int(year_to) if str(year_to or '').isdigit() else None
for raw in results or []:
r = _normalize_result_links(raw)
yr = _parse_year_value(r.get('date'))
if y_from and yr and yr < y_from:
continue
if y_to and yr and yr > y_to:
continue
if not _is_article_like(r):
continue
if not (r.get('title') and (r.get('primo_url') or r.get('link') or r.get('doi'))):
continue
score = 0
if r.get('_source') == 'PRIMO':
score += 4
if r.get('description'):
score += 3
if r.get('doi'):
score += 1
if r.get('link'):
score += 1
if yr:
score += 1
candidates.append((score, r))
candidates.sort(key=lambda item: item[0], reverse=True)
if not candidates or candidates[0][0] < 5:
return None
return candidates[0][1]
async def _summarize_verified_article(question: str, article: dict, behavior: str, use_claude: bool) -> str:
title = article.get('title', 'Untitled')
creators = article.get('creator', 'Unknown')
year = article.get('date', '')
source = article.get('source', '')
primo_url = article.get('primo_url') or _build_primo_search_link(title, article.get('record_id')) or 'Not available'
direct_url = article.get('link') or ('https://doi.org/' + article['doi'] if article.get('doi') else 'Not available')
abstract = (article.get('description') or '').strip()
def _compose(sections: dict | None = None) -> str:
sections = sections or {}
lines = [
'Recommended article',
f'{title} ({year})' if year else title,
'',
'Authors',
creators,
'',
'Source',
source or 'Not clearly stated in the retrieved metadata.',
'',
'Open in PRIMO',
primo_url,
]
if direct_url and direct_url != 'Not available':
lines += ['', 'Direct link', direct_url]
lines += [
'',
'Verification note',
'This answer is grounded only in the retrieved metadata and abstract/description. It does not assume access to the full text unless a direct full-text link is shown.',
'',
'Main research question',
sections.get('Main research question') or (abstract or 'Not clearly stated in the retrieved abstract/metadata.'),
'',
'Methodology',
sections.get('Methodology') or 'Not clearly stated in the retrieved abstract/metadata.',
'',
'Key findings',
sections.get('Key findings') or (abstract or 'Not clearly stated in the retrieved abstract/metadata.'),
'',
'Strengths of the evidence',
sections.get('Strengths of the evidence') or 'Not clearly stated in the retrieved abstract/metadata.',
'',
'Limitations of the evidence',
sections.get('Limitations of the evidence') or 'Not clearly stated in the retrieved abstract/metadata.',
'',
'Contribution to current understanding',
sections.get('Contribution to current understanding') or 'Not clearly stated in the retrieved abstract/metadata.',
]
return '\n'.join(lines)
if not abstract:
return _compose()
prompt = (
f"{behavior}\n\n"
'You are LibBee, the Khalifa University Library AI Assistant. '
'Use ONLY the retrieved abstract or description below. '
'Do NOT invent facts, titles, authors, links, journals, or years. '
'If a section is not explicit, write exactly: Not clearly stated in the retrieved abstract/metadata.\n\n'
'Return ONLY valid JSON with exactly these keys:\n'
'{"Main research question":"...","Methodology":"...","Key findings":"...","Strengths of the evidence":"...","Limitations of the evidence":"...","Contribution to current understanding":"..."}\n\n'
f"User request: {question}\n"
f"Retrieved abstract/description: {abstract}\n"
)
try:
if use_claude:
from langchain_anthropic import ChatAnthropic
llm = ChatAnthropic(model='claude-haiku-4-5-20251001', temperature=0, max_tokens=500)
else:
llm = ChatOpenAI(model='gpt-4o-mini', temperature=0, max_tokens=500)
raw = llm.invoke(prompt).content.strip()
data = _extract_json_object(raw)
allowed = {
'Main research question', 'Methodology', 'Key findings',
'Strengths of the evidence', 'Limitations of the evidence',
'Contribution to current understanding'
}
sections = {k: str(v).strip() for k, v in data.items() if k in allowed and str(v).strip()}
return _compose(sections)
except Exception:
return _compose()
async def _build_search_plan(query: str, year: int = 2026) -> dict:
raw_query = (query or "").strip()
light_query = _light_strip_retrieval_boilerplate(raw_query) or raw_query
prompt = f"""You are a search expert for Khalifa University Library.
The user typed a research request that may contain spelling issues or conversational phrasing.
Create THREE forms while preserving the user's full intent.
1. corrected: lightly edit the FULL request for spelling, grammar, and clarity.
Preserve every substantive concept, constraint, and task.
Do NOT collapse it into a short topic.
2. natural: write a natural-language research query for AI tools such as Consensus,
Perplexity, Semantic Scholar, Scopus AI, PRIMO AI, and LeapSpace.
Keep the full context and constraints, but you may remove only retrieval boilerplate
such as "find an article", "show me", "get me", or "can you find".
Keep constraints like one article vs many, peer reviewed, last five years,
summarize, methodology, findings, strengths, limitations, and contribution.
3. boolean: write a PRIMO/PubMed Boolean query with AND/OR/parentheses that preserves
the main topical concepts and important search constraints.
Examples:
Input: "Find one peer-reviewed article from the last five years on climate change and biodiversity. Summarize the methodology and findings."
→ corrected:"Find one peer-reviewed article from the last five years on climate change and biodiversity. Summarize the methodology and findings."
→ natural:"Peer-reviewed article from the last five years on climate change and biodiversity. Summarize the methodology and findings."
→ boolean:("climate change" OR "global warming") AND (biodiversity OR ecosystems)
Input: "impuct glubal waming on biodiversty"
→ corrected:"impact of global warming on biodiversity"
→ natural:"impact of global warming on biodiversity"
→ boolean:("global warming" OR "climate change") AND biodiversity
Return ONLY valid JSON:
{{"corrected":"full polished request","natural":"full natural-language research query","boolean":"(A OR B) AND (C OR D)","year_from":"","year_to":"","peer_reviewed":false,"open_access":false}}
Query: "{light_query}"""
try:
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0, max_tokens=300)
response = llm.invoke(prompt)
result = _extract_json_object(response.content.strip())
corrected = result.get("corrected", query).strip() or query
natural = result.get("natural", corrected).strip() or corrected
boolean = (result.get("boolean") or "").strip()
has_operators = bool(re.search(r'\b(AND|OR)\b', boolean)) if boolean else False
has_parens = "(" in boolean if boolean else False
if not has_operators or not has_parens:
boolean = _make_boolean(corrected)
return {
"corrected": corrected,
"natural": natural,
"boolean": boolean,
"database_query": _clean_database_keywords(boolean),
"year_from": result.get("year_from", "") or "",
"year_to": result.get("year_to", "") or "",
"peer_reviewed": bool(result.get("peer_reviewed", False)),
"open_access": bool(result.get("open_access", False)),
}
except Exception:
boolean = _make_boolean(raw_query)
corrected = raw_query.strip() or raw_query
natural = _light_strip_retrieval_boilerplate(raw_query) or corrected
return {
"corrected": corrected,
"natural": natural,
"boolean": boolean,
"database_query": _clean_database_keywords(boolean),
"year_from": "",
"year_to": "",
"peer_reviewed": False,
"open_access": False,
}
def get_behavior_instructions() -> str:
base = get_config("bot_personality", "You are a helpful, friendly, and knowledgeable library assistant at Khalifa University, Abu Dhabi, UAE. KU = Khalifa University, NOT Kuwait University. Be concise (2-4 sentences). Always include relevant URLs.")
extra = get_config("custom_instructions", "")
return (base + ("\n" + extra if extra else "")).strip()
def _looks_social_or_greeting(question: str) -> bool:
# SOCIAL_RE is anchored with ^ so .match() is the correct call.
# Using .search() with ^ worked most of the time but could silently
# fail when the engine didn't treat ^ as start-of-string in search mode
# with certain Unicode inputs — .match() is unambiguous.
return bool(SOCIAL_RE.match((question or "").strip()))
async def _interpret_semantics(question: str, history=None) -> dict:
q = (question or "").strip()
ql = q.lower()
canonical_terms = []
grounding_keys = []
social = False
def add(key: str, *terms: str):
if key and key not in grounding_keys:
grounding_keys.append(key)
for t in terms:
if t and t not in canonical_terms:
canonical_terms.append(t)
if _looks_social_or_greeting(q):
social = True
return {"intent_hint": "general", "canonical_terms": canonical_terms, "grounding_keys": grounding_keys, "social": True}
if _looks_library_hours_question(q):
add("circulation", "library hours", "opening hours", "closing hours")
return {"intent_hint": "library_info", "canonical_terms": canonical_terms, "grounding_keys": grounding_keys, "social": False}
if _looks_nonlibrary_ku_question(q):
return {"intent_hint": "general", "canonical_terms": canonical_terms, "grounding_keys": grounding_keys, "social": False}
contact_or_support = bool(re.search(r"\b(who handles|who can help|who should i contact|contact for|email for|phone for|librarian for|help with access|access problem|login issue|remote access problem|vendor issue|technical issue|support)\b", ql))
resource_or_search_task = bool(re.search(r"\b(best|which|recommend|suggest|compare|difference|find|search|articles?|papers?|books?|literature|study|studies|review|summariz(?:e|ing)|summaris(?:e|ing)|evaluate|critique|one article|single article|latest|recent|last \d+ years?)\b", ql))
# Staff / role semantics
if re.search(r"\b(systems? librarian|system librarian|website|digital services|library systems?|technology help)\b", ql) and contact_or_support:
add("systems_help", "Walter Brian Hall", "Systems Librarian", "website", "technology")
if re.search(r"\b(database access|e-?resources?|remote access|off campus|off-campus|login issue|access problem|vendor issue)\b", ql):
if contact_or_support or re.search(r"\b(access problem|login issue|vendor issue|remote access problem|off campus access)\b", ql):
add("database_access", "Rani Anand", "E-Resources", "database access")
else:
add("database_access", "databases", "e-resources", "remote access")
if re.search(r"\b(orcid|open access|apc|article processing charge|research impact|bibliometric|bibliometrics|scival|scopus metrics?)\b", ql):
if contact_or_support and not resource_or_search_task:
add("orcid_oa", "Walter Brian Hall", "ORCID", "Open Access", "APC", "openaccess@ku.ac.ae")
else:
add("orcid_oa", "ORCID", "Open Access", "APC", "research impact")
if re.search(r"\b(research support|research impact|bibliometrics|scival|khazna|scholarly communication|libguides)\b", ql):
if contact_or_support and not resource_or_search_task:
add("research_help", "Nikesh Narayanan", "research support", "bibliometrics", "Khazna")
else:
add("research_help", "research support", "bibliometrics", "Khazna")
if re.search(r"\b(medical librarian|pubmed help|embase|cinahl|cochrane|uptodate|systematic review|clinical databases?)\b", ql):
if contact_or_support and not resource_or_search_task:
add("medical_help", "Jason Fetty", "Medical Librarian", "PubMed", "systematic review")
else:
add("medical_help", "PubMed", "Embase", "CINAHL", "Cochrane", "UpToDate")
if re.search(r"\b(acquisitions?|collection development|suggest a book|request a title|new title request|purchase request|book request)\b", ql):
add("acquisitions", "Alia Al-Harrasi", "Meera Alnaqbi", "Acquisitions", "collection development")
if re.search(r"\b(catalogu(?:e|ing)|cataloging|metadata|cataloguer)\b", ql):
add("cataloguing", "Shamsa Alhosani", "William Hutchinson", "cataloguing", "cataloging", "metadata")
if re.search(r"\b(circulation|borrow|renew|due date|fine|hold request|my library account|public services?)\b", ql):
add("circulation", "borrowing", "renewal", "My Library Account")
if re.search(r"\b(ask a librarian|research help|research consultation|reference help|consultation)\b", ql):
add("research_help", "Ask a Librarian", "research consultation")
# Service / resource semantics
if re.search(r"\b(inter\s*library loan|interlibrary loan|\bill\b|document delivery|full text unavailable|no full text|can't get (an )?article|cannot get (an )?article|article not available)\b", ql):
add("ill", "Suaad Al Jneibi", "Interlibrary Loan", "ILL", "document delivery")
add("fulltext", "full text", "LibKey Nomad", "article access")
if re.search(r"\b(primo|catalog|discovery|holdings|publication finder|does the library have)\b", ql):
add("primo", "PRIMO", "catalog", "Publication Finder")
# Database recommendation semantics
if re.search(r"\b(standards?|astm|ieee standards?|asme|asce)\b", ql):
add("standards", "ASTM Compass", "IEEE standards", "ASME", "ASCE")
if re.search(r"\b(dissertations?|theses|khazna)\b", ql):
add("theses", "ProQuest Dissertations & Theses", "Khazna")
if re.search(r"\b(impact factor|journal metrics?|jcr|citescore|sjr)\b", ql):
add("metrics", "Journal Citation Reports", "CiteScore")
if re.search(r"\b(chemistry|chemical)\b", ql):
add("chemistry", "ACS", "SciFindern", "Reaxys")
if re.search(r"\b(physics|optics?|photonics?)\b", ql):
add("physics", "APS", "AIP", "IOPScience")
if re.search(r"\b(engineering|civil engineering|mechanical engineering|electrical engineering|computer science|ai)\b", ql):
add("engineering", "IEEE Xplore", "ACM", "Knovel", "ASCE", "ASME")
# Name/title style library questions
if re.search(r"\b(who is|who handles|who can help|whom should i contact|contact for)\b", ql) and (_looks_library_question(q) or grounding_keys):
if not grounding_keys:
add("research_help", "staff directory", "library staff")
# Decide intent
if _looks_medical_search(q):
intent = "search_medical"
elif _looks_research_question(q):
intent = "search_academic"
elif grounding_keys or _looks_library_question(q):
intent = "library_info"
elif _looks_current_question(q):
intent = "general_recent"
else:
intent = "general"
return {
"intent_hint": intent,
"canonical_terms": canonical_terms,
"grounding_keys": grounding_keys,
"social": social,
}
def _looks_library_question(question: str) -> bool:
return bool(LIBRARY_CUE_RE.search(question))
def _looks_current_question(question: str) -> bool:
return bool(CURRENT_INFO_RE.search(question))
def _looks_research_question(question: str) -> bool:
q = question.lower()
if RESEARCH_CUE_RE.search(q):
return True
# Natural academic questions that don't use explicit search trigger words
# e.g. "What are the recent advances in AI for cancer diagnosis?"
# e.g. "How effective are mRNA vaccines in treating lung cancer?"
# e.g. "What is the impact of climate change on water security in the UAE?"
academic_patterns = re.compile(
r"\b(impact of|effect[s]? of|role of|use of|application of|advance[s] in|"
r"development[s] in|progress in|challenge[s] of|benefit[s] of|"
r"causes? of|relationship between|comparison of|effectiveness of|"
r"how (effective|efficient|useful|does|do|can|is|are)|"
r"what (are the|is the) (recent|latest|current|new|emerging)|"
r"published in the last|in the last \d+ years?|"
r"systematic review|meta.?analysis|clinical trial[s]?|"
r"literature on|survey of|overview of|state of the art)\b",
re.IGNORECASE
)
return bool(academic_patterns.search(q))
def _looks_medical_search(question: str) -> bool:
q = question.lower()
if not is_medical_topic(q):
return False
return bool(MEDICAL_SEARCH_RE.search(q) or _looks_research_question(q))
def _looks_library_hours_question(question: str) -> bool:
q = question.strip().lower()
# Must match hours pattern AND have at least one library/time/open signal
return bool(HOURS_RE.search(q)) and any(
kw in q for kw in (
'library', 'campus', 'habshan', 'san', 'hours', 'open', 'close',
'schedule', 'till', 'until', 'today', 'tomorrow', 'friday',
'saturday', 'sunday', 'week', 'ramadan'
)
)
def _looks_campus_question(question: str) -> bool:
"""Returns True for questions about KU library locations/branches/campuses."""
return bool(KU_CAMPUS_RE.search(question))
def _looks_nonlibrary_ku_question(question: str) -> bool:
q = question.strip().lower()
if 'library' in q or 'librarian' in q or 'primo' in q or 'database' in q or 'book' in q or 'article' in q:
return False
return bool(KU_GENERAL_RE.search(question))
def _hours_redirect_answer() -> str:
"""
Hardcoded library hours + always direct to the hours page.
Hours are from the KB (ku_library_san_campus.txt). Always advise to check
the link since times can change for exams, holidays, and Ramadan.
"""
return (
"Here are the current <strong>Khalifa University Library</strong> hours:<br><br>"
"<strong>Habshan Library (SAN Campus):</strong><br>"
"Sunday – Wednesday: 8:00 AM – 7:00 PM<br>"
"Thursday: 8:00 AM – 4:00 PM<br>"
"Friday: Closed<br>"
"Saturday: 12:00 PM – 8:00 PM<br><br>"
"⚠️ Hours may change during exam periods, public holidays, and Ramadan. "
"Always check the official Library Hours page for the latest times:<br>"
f"<a href=\"{LIBRARY_HOURS_URL}\" target=\"_blank\">{LIBRARY_HOURS_URL}</a>"
)
def _ku_campus_answer() -> str:
"""
Hardcoded answer for any question about KU library locations/campuses.
Never answered from web — always this fixed text.
"""
return (
"Khalifa University Library has <strong>two branches</strong>:<br><br>"
"<strong>1. Main Campus Library</strong><br>"
"Location: E Building, Main Campus, Abu Dhabi<br>"
"Phone: <a href=\"tel:+97123124604\">+971 2 312 4604</a><br><br>"
"<strong>2. Habshan Library (SAN Campus)</strong><br>"
"Location: Building 5, Ground Floor & First Floor, Sas Al Nakhl (SAN) Campus, Abu Dhabi<br>"
"Phone: <a href=\"tel:+97123123160\">+971 2 312 3160</a><br><br>"
"For general library enquiries: "
"<a href=\"https://library.ku.ac.ae/AskUs\" target=\"_blank\">Ask a Librarian</a> "
"or email <a href=\"mailto:libse@ku.ac.ae\">libse@ku.ac.ae</a>."
)
def _ku_general_redirect_answer() -> str:
return (
f"I’m <strong>LibBee</strong>, the Khalifa University <strong>Library</strong> AI Assistant, so I’m best for library resources, services, databases, research help, and staff contacts.<br><br>"
f"For general Khalifa University questions, please visit the main KU website: <a href=\"{KU_MAIN_URL}\" target=\"_blank\">{KU_MAIN_URL}</a>. "
"A broader university chatbot is available there for non-library questions."
)
@app.post("/correct")
async def correct_query(req: CorrectRequest):
"""
Build three query forms from the user's full request:
- corrected: polished full request
- natural: full natural-language AI-tool query
- boolean: database-ready Boolean query
Only light cleanup is applied before planning.
"""
raw = req.query.strip()
query_to_use = _light_strip_retrieval_boilerplate(raw) or raw
plan = await _build_search_plan(query_to_use, req.year)
return {
"corrected": plan["corrected"],
"boolean": plan["boolean"],
"natural": plan["natural"],
"database_query": plan["database_query"],
"year_from": plan["year_from"],
"year_to": plan["year_to"],
"peer_reviewed": plan["peer_reviewed"],
"open_access": plan["open_access"],
}
# ---- General knowledge endpoint (uses HF Space OpenAI key — separate quota from Cloudflare) ----
class GeneralRequest(BaseModel):
question: str
history: list = []
model: str = "gpt"
async def _answer_general(question: str, history=None) -> dict:
"""Answer general knowledge questions using live web search when available."""
if _looks_library_hours_question(question):
return {"answer": _hours_redirect_answer(), "sources": [], "model": "library-hours-redirect"}
if _looks_nonlibrary_ku_question(question):
return {"answer": _ku_general_redirect_answer(), "sources": [], "model": "ku-general-redirect"}
if _looks_social_or_greeting(question):
if PURE_GREETING_RE.match(question.strip()):
guided = (
"Hi! I’m <strong>LibBee</strong>, the Khalifa University Library AI Assistant.<br><br>"
"I can help you with articles and books, databases, full text, Interlibrary Loan (ILL), "
"library services, staff contacts, ORCID, RefWorks, and Open Access.<br><br>"
"Are you looking for one of these? You can also type your question directly."
)
return {"answer": guided, "sources": [], "model": "libbee-greeting"}
try:
behavior = get_behavior_instructions()
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.4, max_tokens=220)
msgs = []
if behavior:
msgs.append({"role": "system", "content": behavior + "\nFor social, greeting, or self-description questions, answer as LibBee in a warm, brief, natural way. Make it clear that you are LibBee, the Khalifa University Library AI Assistant, and that you help with books, articles, databases, library services, research support, and staff contacts. For questions like 'how are you', 'who are you', 'what are you', 'what can you do', 'are you a bot', playful teasing, or mild mockery, respond politely and briefly, without sounding defensive. If the question is not library-related, you may gently mention that you are specialized in library help. Match the language of the user question."})
for m in (history or [])[-4:]:
if m.get("role") in ("user", "assistant"):
msgs.append({"role": m["role"], "content": m.get("content", "")})
msgs.append({"role": "user", "content": question})
response = llm.invoke(msgs)
return {"answer": response.content.strip(), "sources": [], "model": "gpt-4o-mini-social"}
except Exception:
return {"answer": "I’m <strong>LibBee</strong>, the Khalifa University Library chatbot. I’m here to help with library questions, books, articles, databases, full text, and library services. What would you like help with?", "sources": [], "model": "fallback-social"}
try:
import openai
client = openai.OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
messages = []
if history:
for m in history[-5:]:
role = m.get("role", "user")
msg_content = m.get("content", "")
if role in ("user", "assistant") and msg_content:
messages.append({"role": role, "content": msg_content})
messages.append({"role": "user", "content": question})
response = client.responses.create(
model="gpt-4o-mini",
tools=[{"type": "web_search_preview"}],
input=messages,
)
answer = ""
sources = []
for block in response.output:
if hasattr(block, "content"):
for item in block.content:
if hasattr(item, "text"):
answer += item.text
if hasattr(item, "annotations"):
for ann in item.annotations:
if hasattr(ann, "url") and hasattr(ann, "title"):
sources.append({"url": ann.url, "title": ann.title})
if not answer:
answer = response.output_text if hasattr(response, "output_text") else ""
return {"answer": answer.strip(), "sources": sources, "model": "gpt-4o-mini-web"}
except Exception as e:
print(f"Web search failed, falling back to plain GPT: {e}")
try:
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.3, max_tokens=500)
msgs = []
for m in (history or [])[-5:]:
if m.get("role") in ("user", "assistant"):
msgs.append({"role": m["role"], "content": m.get("content", "")})
msgs.append({"role": "user", "content": question})
response = llm.invoke(msgs)
return {"answer": response.content.strip(), "sources": [], "model": "gpt-4o-mini"}
except Exception as e2:
return {"answer": "", "sources": [], "error": str(e2)}
@app.post("/general")
async def general_query(req: GeneralRequest):
return await _answer_general(req.question, req.history)
# ---- Individual search endpoints ----
@app.post("/search")
async def search(req: SearchRequest):
start = time.time()
source = req.source.lower()
result = {}
try:
if source == "primo":
result = await tool_search_primo(req.query, req.limit, req.peer_reviewed, req.open_access, req.year_from, req.year_to)
elif source == "pubmed":
result = await tool_search_pubmed(req.query, req.limit)
elif source == "scholar":
result = await tool_search_scholar(req.query, req.limit)
elif source == "consensus":
result = await tool_search_consensus(req.query, req.limit)
elif source == "all":
import asyncio
tasks = [
tool_search_primo(req.query, req.limit, req.peer_reviewed, req.open_access, req.year_from, req.year_to),
tool_search_pubmed(req.query, min(req.limit, 3)),
tool_search_scholar(req.query, min(req.limit, 3)),
]
results = await asyncio.gather(*tasks, return_exceptions=True)
combined = []
for r in results:
if isinstance(r, dict) and "results" in r:
for item in r["results"]:
item["_source"] = r.get("source", "unknown")
combined.extend(r["results"])
result = {"total": len(combined), "results": combined, "source": "Multiple"}
else:
result = {"error": f"Unknown source: {source}", "results": []}
elapsed = time.time() - start
# Logging handled by Cloudflare D1 via worker /log endpoint (single source of truth)
result["response_time"] = round(elapsed, 2)
result["is_medical"] = is_medical_topic(req.query)
return result
except Exception as e:
elapsed = time.time() - start
return {"error": str(e), "results": [], "response_time": round(elapsed, 2)}
# ---- Public log endpoint (frontend calls this for all queries) ----
class LogRequest(BaseModel):
question: str
tool: str = "unknown"
model: str = "gpt"
response_time: float = 0
result_count: int = 0
error: str | None = None
@app.post("/log")
async def log_endpoint(req: LogRequest):
log_query(req.question, req.tool, req.model, req.response_time, req.result_count, req.error)
return {"status": "ok"}
# ---- RAG endpoint (backward compatible) ----
@app.post("/rag")
async def rag_query(req: RAGRequest):
start = time.time()
try:
history = [{"role": m.role, "content": m.content} for m in req.history] if req.history else None
result = await tool_library_info(req.question, history, model=req.model)
elapsed = time.time() - start
# Logging handled by Cloudflare D1 via worker /log endpoint
return {
"answer": result["answer"],
"sources": result["sources"],
"has_answer": result.get("has_answer", True),
"model_used": req.model,
"response_time": round(elapsed, 2),
}
except Exception as e:
elapsed = time.time() - start
return {"answer": "", "sources": [], "has_answer": False, "error": str(e)}
# ===== LLM FIRST-GATE CLASSIFIER =====
#
# Architecture (executed in this order for every /agent request):
#
# Layer 0 — INSTANT free responses (no LLM, <1ms)
# • Pure greetings → hardcoded LibBee greeting
# • Exact social phrases → _fixed_social_answer()
# • Staff name match → staff directory
# • Staff role match → staff directory
# • Greeting followup → clarify menu
# • Library hours → hours redirect
# • Non-library KU Q → KU website redirect
#
# Layer 1 — LLM CLASSIFIER (GPT-4o-mini, ~300ms, ~$0.000045/call)
# ALL other messages hit this exactly once.
# Returns: social | library_info | search_academic |
# search_medical | general_recent | general
# + optional casual_answer string for social intent
#
# Layer 2 — EXECUTE
# social → _libbee_casual_response()
# library_info → RAG pipeline (_interpret_semantics + tool_library_info)
# search_* → PRIMO + PubMed/Scholar + RAG synthesis
# general_recent → _answer_general (web search)
# general → _answer_general
_CLASSIFIER_SYSTEM = """You are the routing brain of LibBee, the Khalifa University Library AI Assistant (Abu Dhabi, UAE).
Your only job: read the user message and return a single JSON object routing it to the right handler.
INTENTS — pick exactly one:
"social" — casual chat, greetings, emotions, compliments, complaints, vague help requests
("can you help me", "I'm frustrated", "you're great", "I give up", "help please",
"I don't know what to do", "I am fed up", "I need some guidance", "are you there",
jokes, small talk, anything with no library/research/academic content)
"library_info" — questions about KU Library services, hours, rooms, borrowing, fines, ILL, staff,
databases, ORCID, RefWorks, open access, APC, account, printing, remote access,
LibKey, acquisitions, Khazna, thesis submission, library policies, alumni access
"search_academic"— user wants to FIND articles / papers / books / literature on a specific academic topic
(e.g. "find articles on renewable energy", "papers on machine learning in healthcare",
"I need books about civil engineering", "literature review on AI ethics")
"search_medical" — same as search_academic but topic is medical/clinical/nursing/pharmacy/health sciences
"general_recent" — questions about current events, news, latest info, who holds a position now,
real-time data (stock prices, today's weather, breaking news)
"general" — factual/knowledge question NOT about KU Library and NOT a research search
(explain a concept, historical facts, definitions, "what is quantum computing")
RULES:
1. When in doubt between library_info and social, choose library_info if ANY library keyword is present.
2. "I don't know how to search for articles" = library_info (has library task content).
3. "I am fed up with searching" alone (no specific topic/database) = social (expressing frustration).
4. "I need help finding articles on diabetes" = search_medical.
5. KU = Khalifa University. Do NOT confuse with Kuwait University.
6. ANY question asking about advances, approaches, methods, effectiveness, impact, or relationship
on a specific academic/scientific topic = search_academic or search_medical. NOT library_info.
e.g. "What are the recent advances in AI for predictive maintenance?" = search_academic
e.g. "How effective are mRNA vaccines for cancer treatment?" = search_medical
e.g. "What is the impact of climate change on food security?" = search_academic
For "social" intent ONLY: also include "casual_answer" — a warm 1-3 sentence LibBee response
(friendly librarian tone, no markdown, no bullet points). Offer to help with library services.
For all other intents: omit "casual_answer" entirely.
Respond ONLY with valid JSON, no preamble, no markdown fences.
Examples:
{"intent":"social","casual_answer":"Of course, I\'m here to help! I\'m LibBee, the KU Library AI Assistant. What would you like to find today?"}
{"intent":"social","casual_answer":"I completely understand that feeling — searching can be tricky at first. I\'m LibBee, the Khalifa University Library AI Assistant, and I can guide you through it step by step. What topic are you trying to research?"}
{"intent":"library_info"}
{"intent":"search_academic"}
{"intent":"search_medical"}
{"intent":"general_recent"}
{"intent":"general"}"""
async def _llm_classify(question: str, history: list) -> dict:
"""
Single LLM call that classifies intent for ALL non-trivial messages.
~300ms latency, ~$0.000045 per call at GPT-4o-mini pricing.
Returns: {"intent": str, "casual_answer": str}
Never raises — falls back to {"intent":"general","casual_answer":""} on error.
"""
# Build compact conversation context (last 2 turns, 100 chars each)
ctx_parts = []
for m in history[-4:]:
role = m.get("role", "")
text = (m.get("content") or "")[:100]
if role in ("user", "assistant") and text:
ctx_parts.append(f"{role}: {text}")
ctx = " | ".join(ctx_parts) if ctx_parts else "none"
user_msg = (
"Message: " + question[:400] + "\n" + "Context: " + ctx + "\nJSON:"
)
try:
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0, max_tokens=150)
response = llm.invoke([
{"role": "system", "content": _CLASSIFIER_SYSTEM},
{"role": "user", "content": user_msg},
])
raw = response.content.strip()
if raw.startswith("```"):
raw = raw.split("\n", 1)[1].rsplit("```", 1)[0].strip()
result = json.loads(raw)
intent = result.get("intent", "general")
valid_intents = {"social","library_info","search_academic","search_medical","general_recent","general"}
if intent not in valid_intents:
intent = "general"
return {"intent": intent, "casual_answer": result.get("casual_answer", "")}
except Exception as e:
print(f"[LibBee classifier error] {e} | question: {question[:80]}")
return {"intent": "general", "casual_answer": ""}
async def _libbee_casual_response(question: str, history: list, model: str, hint: str = "") -> str:
"""
Generate a warm, in-character LibBee response for social/casual messages.
Uses the classifier hint when available (saves a second LLM call).
Falls back to a hardcoded response on any error.
"""
# Use the classifier's casual_answer hint if it's good enough
if hint and len(hint.strip()) > 25:
return hint.strip()
behavior = get_behavior_instructions()
sys_prompt = (
f"{behavior}\n\n"
"You are LibBee, the Khalifa University Library AI Assistant. "
"The user has sent a casual, social, or emotional message. "
"Respond warmly and naturally in 1-3 sentences as a friendly, knowledgeable librarian. "
"Always identify yourself as LibBee and offer library help. "
"If the user seems frustrated or confused, be empathetic and reassuring. "
"If the user seems happy or is complimenting you, be cheerful and grateful. "
"Never say \'I don\'t have information\'. Never use bullet points or markdown. "
"Match the user\'s energy — if casual, be casual; if distressed, be warm and calm."
)
msgs = [{"role": "system", "content": sys_prompt}]
for m in history[-4:]:
if m.get("role") in ("user", "assistant"):
msgs.append({"role": m["role"], "content": (m.get("content") or "")[:200]})
msgs.append({"role": "user", "content": question})
use_claude = model == "claude" and os.environ.get("ANTHROPIC_API_KEY")
try:
if use_claude:
from langchain_anthropic import ChatAnthropic
llm = ChatAnthropic(model="claude-haiku-4-5-20251001", temperature=0.6, max_tokens=200)
else:
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.6, max_tokens=200)
return llm.invoke(msgs).content.strip()
except Exception:
return (
"I'm <strong>LibBee</strong>, the Khalifa University Library AI Assistant — "
"always happy to help! What can I find for you today?"
)
def _make_agent_response(answer, intent, tools_used, search_results, sources,
model, elapsed, question, natural_query=None,
database_query=None, original_question=None,
is_follow_up=False, source_mode="", ask_librarian=None):
"""
Helper to build the standard /agent response dict.
ask_librarian controls whether the frontend appends the
"Ask a Librarian" footer block:
None → frontend decides based on intent + tools_used (default behaviour)
True → always show footer
False → never show footer (e.g. hours/campus answers are self-contained)
"""
return {
"answer": answer,
"intent": intent,
"tools_used": tools_used or [],
"search_results": search_results or [],
"sources": sources or [],
"model_used": model,
"response_time": round(elapsed, 2),
"corrected_query": question,
"natural_query": natural_query or question,
"database_query": database_query or question,
"original_question": original_question or question,
"is_follow_up": is_follow_up,
"source_mode": source_mode,
"ask_librarian": ask_librarian, # None | True | False
}
# ---- Agent endpoint (Batch 3: tool-calling agent) ----
@app.post("/agent")
async def agent_query(req: AgentRequest):
"""
LLM-first multi-tool agent.
Execution order:
Layer 0: Instant free handlers (pure greetings, exact social, staff, hours, KU redirect)
Layer 1: LLM classifier → one GPT-4o-mini call for ALL other messages
Layer 2: Execute the classified intent (social/library/search/general)
"""
start = time.time()
question = req.question.strip()
history = [{"role": m.role, "content": m.content} for m in req.history] if req.history else []
use_claude = req.model == "claude" and bool(os.environ.get("ANTHROPIC_API_KEY"))
q_clean = question.strip()
# ══════════════════════════════════════════════════════════════════
# LAYER 0 — INSTANT FREE HANDLERS (no LLM call, < 1ms each)
# Only for responses that are 100% deterministic and zero-cost.
# ══════════════════════════════════════════════════════════════════
# 0a. Pure greetings (hi / hello / hey / good morning …)
if PURE_GREETING_RE.match(q_clean):
elapsed = time.time() - start
return _make_agent_response(
answer=(
"Hi! I'm <strong>LibBee</strong>, the Khalifa University Library AI Assistant.<br><br>"
"I can help you find articles and books, search databases, access full text, "
"request Interlibrary Loan (ILL), and answer questions about library services and staff.<br><br>"
"Are you looking for one of these? You can also type your question directly."
),
intent="social_greeting", tools_used=[], search_results=[], sources=[],
model=req.model, elapsed=elapsed, question=question,
is_follow_up=False, source_mode="social",
)
# 0b. Exact social phrases (who are you, how are you, thanks, bye …)
fixed = _fixed_social_answer(q_clean)
if fixed:
elapsed = time.time() - start
return _make_agent_response(
answer=fixed, intent="social_greeting",
tools_used=[], search_results=[], sources=[],
model=req.model, elapsed=elapsed, question=question,
is_follow_up=False, source_mode="social",
)
# 0c. Staff name direct lookup (e.g. "nikesh", "jason fetty")
staff_match = _match_staff_name(question)
if staff_match:
elapsed = time.time() - start
src_title = staff_match.get("source_title", "")
src_url = staff_match.get("source", "")
return _make_agent_response(
answer=_staff_name_answer(staff_match, question),
intent="library_info", tools_used=["staff_name_match"],
search_results=[],
sources=([{"title": src_title, "source": src_url}] if src_title or src_url else []),
model=req.model, elapsed=elapsed, question=question,
source_mode="staff_kb" if kb_staff_directory else "staff_directory",
)
# 0d. Staff role lookup (e.g. "who is the medical librarian")
staff_role_match = _match_staff_role(question)
if staff_role_match:
elapsed = time.time() - start
src_title = staff_role_match.get("source_title", "")
src_url = staff_role_match.get("source", "")
return _make_agent_response(
answer=_staff_role_answer(staff_role_match, question),
intent="library_info", tools_used=["staff_role_match"],
search_results=[],
sources=([{"title": src_title, "source": src_url}] if src_title or src_url else []),
model=req.model, elapsed=elapsed, question=question,
source_mode="staff_kb" if kb_staff_directory else "staff_directory",
)
# 0e. Greeting menu follow-up (yes / sure / okay after LibBee intro)
if _is_greeting_menu_followup(question, history):
elapsed = time.time() - start
return _make_agent_response(
answer=_greeting_menu_clarify_answer(),
intent="social_greeting", tools_used=[], search_results=[], sources=[],
model=req.model, elapsed=elapsed, question=question,
is_follow_up=True, source_mode="social",
)
# 0f. Library hours question (fast-path — hardcoded hours + link, never web)
if _looks_library_hours_question(question):
elapsed = time.time() - start
return _make_agent_response(
answer=_hours_redirect_answer(),
intent="library_info", tools_used=["hours_hardcoded"],
search_results=[],
sources=[{"title": "Library Hours", "source": LIBRARY_HOURS_URL}],
model=req.model, elapsed=elapsed, question=question,
ask_librarian=False, # hours answer is self-contained
)
# 0f2. Campus / library location question (hardcoded — never from web)
if _looks_campus_question(question):
elapsed = time.time() - start
return _make_agent_response(
answer=_ku_campus_answer(),
intent="library_info", tools_used=["campus_hardcoded"],
search_results=[], sources=[],
model=req.model, elapsed=elapsed, question=question,
ask_librarian=False, # answer already includes contact info
)
# 0g. Non-library KU question (admissions, programs, fees …)
if _looks_nonlibrary_ku_question(question):
elapsed = time.time() - start
return _make_agent_response(
answer=_ku_general_redirect_answer(),
intent="general", tools_used=["ku_general_redirect"],
search_results=[], sources=[],
model=req.model, elapsed=elapsed, question=question,
)
# ══════════════════════════════════════════════════════════════════
# LAYER 1 — LLM CLASSIFIER
# One GPT-4o-mini call classifies ALL messages that reached here.
# Cost: ~$0.000045 per call. At 250 questions/day ≈ $1/month.
# ══════════════════════════════════════════════════════════════════
cls_result = await _llm_classify(question, history)
intent = cls_result.get("intent", "general")
casual_hint = cls_result.get("casual_answer", "")
# ══════════════════════════════════════════════════════════════════
# LAYER 2 — EXECUTE CLASSIFIED INTENT
# ══════════════════════════════════════════════════════════════════
# ── Social / casual ──────────────────────────────────────────────
if intent == "social":
casual = await _libbee_casual_response(question, history, req.model, casual_hint)
elapsed = time.time() - start
return _make_agent_response(
answer=casual, intent="social_greeting",
tools_used=["libbee_casual"], search_results=[], sources=[],
model=req.model, elapsed=elapsed, question=question,
is_follow_up=False, source_mode="social",
)
# ── Library info — RAG pipeline ───────────────────────────────────
if intent == "library_info":
# Use _interpret_semantics to enrich the query with canonical library terms
enriched_question = question
try:
interp = await _interpret_semantics(question, history)
if interp.get("grounding_keys"):
resolved = " ".join(
[question]
+ interp.get("canonical_terms", [])
+ [GROUNDED_LIBRARY_MAP.get(k, "")
for k in interp.get("grounding_keys", [])
if GROUNDED_LIBRARY_MAP.get(k)]
)
enriched_question = re.sub(r"\s+", " ", resolved).strip()
except Exception:
pass
rag = await tool_library_info(enriched_question, history[-5:] if history else None, model=req.model)
tools_used = ["get_library_info"]
if rag.get("has_answer") and rag.get("answer"):
# Good RAG answer — synthesise and return
context_parts = [f"Library Knowledge Base:\n{rag['answer']}"]
behavior = get_behavior_instructions()
synthesis_prompt = (
f"{behavior}\n\n"
"You are the Khalifa University Library AI Assistant (Abu Dhabi, UAE). KU = Khalifa University.\n"
"Answer in 2-4 sentences. Include URLs from the context when relevant.\n"
"Answer ONLY from the provided context. Do not add information not present in the context.\n\n"
f"Context:\n{chr(10).join(context_parts)}\n\n"
f"Question: {question}\nAnswer:"
)
try:
if use_claude:
from langchain_anthropic import ChatAnthropic
synth_llm = ChatAnthropic(model="claude-haiku-4-5-20251001", temperature=0.2, max_tokens=500)
else:
synth_llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.2, max_tokens=500)
answer = synth_llm.invoke(synthesis_prompt).content.strip()
except Exception:
answer = rag["answer"]
elapsed = time.time() - start
return _make_agent_response(
answer=answer, intent="library_info", tools_used=tools_used,
search_results=[], sources=rag.get("sources", []),
model=req.model, elapsed=elapsed, question=question,
ask_librarian=False, # RAG answer from trusted KB — no footer needed
)
else:
# ── RAG miss — three-tier fallback for library questions ──
#
# Tier 1: Strict KU-library-only LLM prompt (no hallucination risk —
# LLM is told to say it doesn't know rather than invent).
# Tier 2: If tier 1 is vague/empty → web search scoped to library.ku.ac.ae
# so all facts come from the real KU library website.
# Tier 3: Always append "Ask a Librarian" + library homepage link.
#
# General web search (_answer_general) is NOT used here — that's only
# for intent==general/general_recent where unrestricted search is correct.
# ── Tier 1: Strict KU-only LLM ──
ku_only_prompt = (
"You are LibBee, the Khalifa University Library AI Assistant (Abu Dhabi, UAE). "
"KU = Khalifa University.\n\n"
"STRICT RULES:\n"
"1. Answer ONLY using your knowledge of Khalifa University Library services, "
"databases, policies, staff, and resources.\n"
"2. Be concise (2-4 sentences). Include real KU library URLs when relevant "
"(e.g. https://library.ku.ac.ae/AskUs, https://library.ku.ac.ae/ill/, "
"https://library.ku.ac.ae/eresources, https://library.ku.ac.ae/oa/).\n"
"3. If you are not confident about a specific detail (phone number, exact policy, "
"specific database URL), do NOT invent it — instead say you are not certain and "
"direct the user to Ask a Librarian or the library website.\n"
"4. Never make up staff names, contact details, or database names.\n"
"5. Do NOT use general web knowledge or facts unrelated to KU Library.\n\n"
f"Question: {question}\nAnswer:"
)
tier1_answer = ""
try:
if use_claude:
from langchain_anthropic import ChatAnthropic
t1_llm = ChatAnthropic(model="claude-haiku-4-5-20251001", temperature=0.1, max_tokens=400)
else:
t1_llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.1, max_tokens=400)
t1_response = t1_llm.invoke([
{"role": "system", "content": ku_only_prompt},
{"role": "user", "content": question},
])
tier1_answer = t1_response.content.strip()
except Exception as t1_err:
print(f"[LibBee tier1 LLM error] {t1_err}")
# ── Tier 2: If tier 1 is vague, search library.ku.ac.ae ──
# Detect vague answers: short, uncertain, or explicitly admitting no knowledge
vague_phrases = [
"i'm not sure", "i am not sure", "i don't have", "i do not have",
"not certain", "cannot confirm", "not available", "unable to find",
"please contact", "you may want to", "i would recommend checking",
"i don't have specific", "i lack specific",
]
is_vague = (
not tier1_answer
or len(tier1_answer.split()) < 15
or any(p in tier1_answer.lower().replace("\u2019", "'") for p in vague_phrases)
)
tier2_answer = ""
tier2_sources = []
if is_vague:
try:
import openai as _openai
_client = _openai.OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
# Scope web search to KU library domain only
scoped_query = f"site:library.ku.ac.ae {question}"
_resp = _client.responses.create(
model="gpt-4o-mini",
tools=[{"type": "web_search_preview"}],
input=[{
"role": "system",
"content": (
"You are LibBee, the Khalifa University Library AI Assistant. "
"Search the KU library website (library.ku.ac.ae) only. "
"Answer in 2-4 sentences using only what you find on the KU library website. "
"Always include the source URL. "
"If nothing relevant is found on library.ku.ac.ae, say so briefly."
)
}, {"role": "user", "content": scoped_query}],
)
for block in _resp.output:
if hasattr(block, "content"):
for item in block.content:
if hasattr(item, "text"):
tier2_answer += item.text
if hasattr(item, "annotations"):
for ann in item.annotations:
if hasattr(ann, "url") and hasattr(ann, "title"):
if "library.ku.ac.ae" in getattr(ann, "url", ""):
tier2_sources.append({"url": ann.url, "title": ann.title})
if not tier2_answer:
tier2_answer = getattr(_resp, "output_text", "") or ""
tier2_answer = tier2_answer.strip()
except Exception as t2_err:
print(f"[LibBee tier2 web search error] {t2_err}")
# ── Combine best answer — ask_librarian=True for all miss tiers ──
# The frontend will render the Ask a Librarian block.
# Answer strings are clean — no embedded HTML footer.
if tier2_answer and len(tier2_answer.split()) >= 10:
# Web search on library.ku.ac.ae found something useful
final_answer = tier2_answer
final_sources = tier2_sources
final_tools = tools_used + ["ku_web_search"]
elif tier1_answer and len(tier1_answer.split()) >= 10:
# Tier 1 KU-only LLM gave a decent answer
final_answer = tier1_answer
final_sources = []
final_tools = tools_used + ["ku_only_llm"]
else:
# Both failed — clean redirect only
final_answer = (
"I don't have that specific information in my knowledge base right now. "
"The KU Library team will be able to help directly."
)
final_sources = []
final_tools = tools_used + ["redirected"]
elapsed = time.time() - start
return _make_agent_response(
answer=final_answer, intent="library_info",
tools_used=final_tools,
search_results=[], sources=final_sources,
model=req.model, elapsed=elapsed, question=question,
ask_librarian=True, # Tier 1/2/3 — always show footer
)
# ── Academic / medical search ─────────────────────────────────────
if intent in ("search_academic", "search_medical"):
import asyncio as _asyncio
search_plan = await _build_search_plan(question)
corrected_query = search_plan["corrected"]
natural_query = search_plan["natural"]
database_query = search_plan["database_query"] or search_plan["corrected"]
year_from = search_plan.get("year_from") or None
year_to = search_plan.get("year_to") or None
peer_reviewed = bool(search_plan.get("peer_reviewed"))
open_access = bool(search_plan.get("open_access"))
tasks = [tool_search_primo(database_query, limit=5, peer_reviewed=peer_reviewed, open_access=open_access, year_from=year_from, year_to=year_to)]
if intent == "search_medical":
tasks.append(tool_search_pubmed(database_query, limit=3))
else:
tasks.append(tool_search_scholar(natural_query, limit=3))
raw_results = await _asyncio.gather(*tasks, return_exceptions=True)
combined = []
tools_used = []
for r in raw_results:
if isinstance(r, dict) and r.get("results"):
source_name = r.get("source", "unknown")
tools_used.append(source_name)
for item in r["results"]:
merged = dict(item)
merged.setdefault("_source", source_name)
combined.append(_normalize_result_links(merged))
wants_single_article = bool(re.search(r"\b(one|single)\s+(peer.?reviewed\s+)?(article|paper|study)\b", question, re.IGNORECASE))
wants_structured_summary = bool(re.search(r"\b(summariz(?:e|ing)|summaris(?:e|ing)|main research question|methodology|methods?|key findings|strengths?|limitations?|critical(?:ly)? evaluate|critique|contribution)\b", question, re.IGNORECASE))
if wants_single_article or wants_structured_summary:
preferred = [r for r in combined if r.get("_source") == "PRIMO"]
secondary = [r for r in combined if r.get("_source") != "PRIMO"]
combined = preferred + secondary
rag = await tool_library_info(question, history[-3:] if history else None, model=req.model)
tools_used.append("get_library_info")
tools_used = list(dict.fromkeys(tools_used))
behavior = get_behavior_instructions()
if wants_single_article or wants_structured_summary:
candidate = _choose_verified_article(combined, year_from=year_from, year_to=year_to)
if candidate:
answer = await _summarize_verified_article(question, candidate, behavior, use_claude)
else:
answer = (
"I couldn’t confidently retrieve one exact article matching your full request from the current results. "
"Please look into the available AI tools below to refine or expand the search."
)
else:
context_parts = []
if rag.get("answer"):
context_parts.append(f"Library Knowledge Base:\n{rag['answer']}")
if combined:
top = combined[:5]
res_lines = []
for idx, r in enumerate(top, 1):
res_lines.append(
f"{idx}. Title: {r.get('title','')}\n"
f" Authors: {r.get('creator','')}\n"
f" Year: {r.get('date','')}\n"
f" Source: {r.get('source','')}\n"
f" Type: {r.get('type','')}\n"
f" DOI: {r.get('doi','')}\n"
f" PRIMO link: {r.get('primo_url','')}\n"
f" Direct link: {r.get('link','')}\n"
f" Abstract/Description: {r.get('description','')}"
)
context_parts.append("Candidate Search Results:\n" + "\n\n".join(res_lines))
context_parts.append(f"Natural query for AI tools: {natural_query}")
context_parts.append(f"Database query for PRIMO/PubMed: {database_query}")
synthesis_prompt = (
f"{behavior}\n\n"
"You are the KU Library AI Assistant. Be concise but helpful (4-7 sentences).\n"
"Answer the user's search request directly, mention the search direction, and mention 1-3 strong retrieved results when present.\n"
"Do not invent titles, links, or findings beyond the retrieved metadata/abstracts.\n"
"Keep the answer in the chat rather than redirecting the user elsewhere.\n\n"
f"Context:\n{chr(10).join(context_parts) if context_parts else 'No additional context.'}\n\n"
f"Question: {question}\nAnswer:"
)
try:
if use_claude:
from langchain_anthropic import ChatAnthropic
synth_llm = ChatAnthropic(model="claude-haiku-4-5-20251001", temperature=0.2, max_tokens=900)
else:
synth_llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.2, max_tokens=900)
answer = synth_llm.invoke(synthesis_prompt).content.strip()
except Exception as ex:
answer = rag.get("answer", f"Error generating answer: {ex}")
elapsed = time.time() - start
return _make_agent_response(
answer=answer, intent=intent, tools_used=tools_used,
search_results=combined[:8], sources=rag.get("sources", []),
model=req.model, elapsed=elapsed, question=question,
natural_query=natural_query, database_query=database_query,
original_question=question,
)
# ── General / general_recent — web search or plain LLM ───────────
general = await _answer_general(question, history)
gen_answer = general.get("answer", "").strip()
# Guard: if general returned a "no info" string, replace with a helpful redirect
no_info_phrases = [
"i don't have", "i do not have", "no specific information",
"not available in my", "unable to find", "cannot find",
]
gen_lower = gen_answer.lower().replace("\u2019", "'")
if not gen_answer or any(p in gen_lower for p in no_info_phrases):
gen_answer = (
"I'm not sure about that one. For library-specific questions, try "
"<a href=\"https://library.ku.ac.ae/AskUs\" target=\"_blank\">Ask a Librarian</a>. "
"For general knowledge questions, you could also try a web search."
)
elapsed = time.time() - start
return _make_agent_response(
answer=gen_answer, intent=intent,
tools_used=["general_web" if intent == "general_recent" else "general"],
search_results=[], sources=general.get("sources", []),
model=req.model, elapsed=elapsed, question=question,
)
# ---- Rebuild index (protected) ----
@app.post("/rebuild")
async def rebuild_index(request: Request):
if not verify_session(request):
raise HTTPException(status_code=401, detail="Not authenticated")
global vectorstore
try:
import shutil, glob as _glob
# Delete old FAISS index files so we rebuild from scratch
if os.path.exists(FAISS_INDEX_PATH):
shutil.rmtree(FAISS_INDEX_PATH)
print(f"Deleted old FAISS index at {FAISS_INDEX_PATH}")
# List all knowledge files found
kb_files = _glob.glob(os.path.join(KNOWLEDGE_DIR, "*.txt"))
print(f"Knowledge files found: {len(kb_files)}")
for f in kb_files:
print(f" - {os.path.basename(f)} ({os.path.getsize(f)} bytes)")
docs = load_documents()
if not docs:
return {"error": "No knowledge files found", "kb_dir": KNOWLEDGE_DIR, "files": []}
# Force rebuild — ignore any cached index
vectorstore = build_vectorstore(docs, force_rebuild=True)
chunks = vectorstore.index.ntotal
return {
"status": "ok",
"chunks": chunks,
"documents": len(docs),
"files": [os.path.basename(f) for f in kb_files],
"kb_dir": KNOWLEDGE_DIR
}
except Exception as e:
import traceback
return {"error": str(e), "traceback": traceback.format_exc()}
# ===== ADMIN LOGIN =====
@app.get("/admin/login", response_class=HTMLResponse)
async def admin_login_page():
return """<!DOCTYPE html>
<html><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1">
<title>KU Library AI — Admin Login</title>
<style>
*{box-sizing:border-box;margin:0;padding:0}
body{font-family:'Segoe UI',system-ui,sans-serif;background:#f0f4f8;display:flex;justify-content:center;align-items:center;min-height:100vh}
.login-box{background:#fff;border-radius:16px;padding:40px;width:360px;box-shadow:0 4px 24px rgba(0,0,0,.1);text-align:center}
h1{color:#003366;font-size:1.3rem;margin-bottom:6px}
.subtitle{color:#6b7280;font-size:.84rem;margin-bottom:24px}
input{width:100%;padding:12px 14px;border:1.5px solid #d1d5db;border-radius:10px;font-size:.92rem;margin-bottom:14px}
input:focus{outline:none;border-color:#003366}
.btn{width:100%;padding:12px;background:#003366;color:#fff;border:none;border-radius:10px;font-weight:700;font-size:.92rem;cursor:pointer}
.btn:hover{background:#004488}
.error{color:#dc2626;font-size:.82rem;margin-bottom:10px;display:none}
</style></head><body>
<div class="login-box">
<h1>🔐 Admin Login</h1>
<div class="subtitle">KU Library AI Dashboard</div>
<div class="error" id="err">Incorrect password. Try again.</div>
<form onsubmit="return doLogin(event)">
<input type="password" id="pw" placeholder="Enter admin password" autofocus>
<button type="submit" class="btn">Login</button>
</form>
</div>
<script>
async function doLogin(e){
e.preventDefault();
const pw=document.getElementById('pw').value;
const r=await fetch('/admin/auth',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({password:pw})});
if(r.ok){window.location.href='/admin';}
else{document.getElementById('err').style.display='block';document.getElementById('pw').value='';document.getElementById('pw').focus();}
return false;
}
</script></body></html>"""
@app.post("/admin/auth")
async def admin_auth(request: Request, response: Response):
data = await request.json()
if data.get("password") == ADMIN_PASSWORD:
token = create_session()
resp = Response(content=json.dumps({"status": "ok"}), media_type="application/json")
resp.set_cookie("admin_session", token, httponly=True, max_age=86400, samesite="lax")
return resp
raise HTTPException(status_code=401, detail="Invalid password")
@app.get("/admin/logout")
async def admin_logout(request: Request):
token = request.cookies.get("admin_session")
if token and token in admin_sessions:
del admin_sessions[token]
resp = RedirectResponse(url="/admin/login")
resp.delete_cookie("admin_session")
return resp
# ===== ADMIN DASHBOARD (protected) =====
WORKER_URL = "https://steep-haze-e1f8.nikeshn.workers.dev"
@app.get("/admin", response_class=HTMLResponse)
async def admin_dashboard(request: Request):
if not verify_session(request):
return RedirectResponse(url="/admin/login")
config = {}
conn = sqlite3.connect(DB_PATH)
conn.row_factory = sqlite3.Row
rows = conn.execute("SELECT key, value FROM bot_config").fetchall()
conn.close()
for r in rows: config[r["key"]] = r["value"]
def esc(s):
return (s or "").replace("&","&").replace("<","<").replace(">",">").replace('"',""")
return f"""<!DOCTYPE html>
<html><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1">
<title>KU Library AI — Admin</title>
<style>
*{{box-sizing:border-box;margin:0;padding:0}}
body{{font-family:'Segoe UI',system-ui,sans-serif;background:#f0f4f8;color:#1a1a2e;padding:20px;max-width:1200px;margin:0 auto}}
h1{{color:#003366;font-size:1.3rem;margin-bottom:4px}}
.sub{{color:#6b7280;font-size:.82rem;margin-bottom:14px}}
.tabs{{display:flex;gap:2px;border-bottom:2px solid #d1d5db;margin-bottom:14px}}
.tab{{padding:8px 16px;cursor:pointer;font-weight:600;font-size:.84rem;color:#6b7280;border:none;background:none;border-bottom:3px solid transparent;margin-bottom:-2px}}
.tab.active{{color:#003366;border-bottom-color:#003366}}
.tc{{display:none}}.tc.active{{display:block}}
.grid{{display:grid;grid-template-columns:repeat(auto-fit,minmax(160px,1fr));gap:10px;margin-bottom:16px}}
.stat{{background:#fff;border-radius:10px;padding:12px;border:1px solid #d1d5db;text-align:center}}
.stat .n{{font-size:1.6rem;font-weight:800;color:#003366}}.stat .l{{font-size:.76rem;color:#6b7280}}
.card{{background:#fff;border-radius:10px;padding:14px;border:1px solid #d1d5db;margin-bottom:12px}}
.card h2{{font-size:.9rem;color:#003366;margin-bottom:8px}}
table{{width:100%;border-collapse:collapse;font-size:.78rem}}
th{{background:#003366;color:#fff;padding:6px 8px;text-align:left}}
td{{padding:5px 8px;border-bottom:1px solid #e5e7eb}}
tr:hover{{background:#f9fafb}}
.badge{{padding:2px 7px;border-radius:8px;font-size:.68rem;font-weight:600}}
.err{{color:#dc2626;font-size:.72rem}}
canvas{{max-height:180px}}
.fg{{margin-bottom:12px}}
.fg label{{display:block;font-weight:700;font-size:.82rem;color:#003366;margin-bottom:3px}}
.fg .h{{font-size:.72rem;color:#6b7280;margin-bottom:3px}}
textarea,input[type=text],select{{width:100%;padding:9px;border:1.5px solid #d1d5db;border-radius:8px;font-family:inherit;font-size:.82rem;resize:vertical}}
textarea:focus,input:focus,select:focus{{outline:none;border-color:#003366}}
.btn{{padding:9px 20px;border:none;border-radius:8px;font-weight:700;cursor:pointer;font-size:.84rem}}
.bp{{background:#003366;color:#fff}}.bp:hover{{background:#004488}}
.bg{{background:#059669;color:#fff}}
.br{{background:#dc2626;color:#fff}}
.toggle{{display:flex;align-items:center;gap:8px}}
.toggle input{{width:18px;height:18px;accent-color:#003366}}
.st{{padding:8px 14px;border-radius:8px;font-size:.8rem;font-weight:600;display:none;margin-top:8px}}
.st.ok{{background:#d1fae5;color:#065f46;display:block}}
.st.er{{background:#fecaca;color:#991b1b;display:block}}
.two{{display:grid;grid-template-columns:1fr 1fr;gap:12px}}
@media(max-width:768px){{.two{{grid-template-columns:1fr}}}}
#loading{{text-align:center;padding:40px;color:#6b7280}}
</style>
<script src="https://cdnjs.cloudflare.com/ajax/libs/Chart.js/4.4.1/chart.umd.min.js"></script>
</head><body>
<h1>📊 KU Library AI — Admin <span style="float:right;display:flex;gap:6px;align-items:center"><button onclick="location.reload()" class="btn bp" style="padding:5px 12px;font-size:.76rem">🔄 Refresh</button><a href="/admin/logout" style="font-size:.76rem;color:#dc2626;text-decoration:none;padding:5px 12px;border:1.5px solid #dc2626;border-radius:8px;font-weight:600">🔒 Logout</a></span></h1>
<div class="sub">Khalifa University Library AI Agent</div>
<div class="tabs">
<div class="tab active" onclick="st(this,'analytics')">📈 Analytics</div>
<div class="tab" onclick="st(this,'behavior')">🤖 Bot Behavior</div>
<div class="tab" onclick="st(this,'queries')">📋 Query Log</div>
<div class="tab" onclick="st(this,'system')">⚙️ System</div>
</div>
<div class="tc active" id="t-analytics"><div id="loading">Loading analytics from Cloudflare D1…</div></div>
<div class="tc" id="t-behavior">
<div class="card"><h2>🤖 Bot Behavior</h2><p style="font-size:.8rem;color:#6b7280;margin-bottom:12px">Changes apply immediately.</p>
<div class="fg"><label>Welcome Message</label><div class="h">First message users see.</div><textarea id="c-welcome_message" rows="3">{esc(config.get('welcome_message',''))}</textarea></div>
<div class="fg"><label>Bot Personality / System Prompt</label><div class="h">Core instructions for every LLM call.</div><textarea id="c-bot_personality" rows="4">{esc(config.get('bot_personality',''))}</textarea></div>
<div class="fg"><label>Custom Instructions</label><div class="h">Extra guidance, e.g. "During Ramadan, mention reduced hours."</div><textarea id="c-custom_instructions" rows="3">{esc(config.get('custom_instructions',''))}</textarea></div>
<div class="fg"><label>📢 Announcement Banner</label><div class="h">Shows at top of chat. Leave empty to hide.</div><textarea id="c-announcement" rows="2">{esc(config.get('announcement',''))}</textarea></div>
<div class="two"><div class="fg"><label>Default Model</label><select id="c-default_model"><option value="gpt" {'selected' if config.get('default_model')=='gpt' else ''}>ChatGPT</option><option value="claude" {'selected' if config.get('default_model')=='claude' else ''}>Claude</option></select></div>
<div class="fg"><label>Max Results</label><select id="c-max_results"><option value="3" {'selected' if config.get('max_results')=='3' else ''}>3</option><option value="5" {'selected' if config.get('max_results')=='5' else ''}>5</option><option value="10" {'selected' if config.get('max_results')=='10' else ''}>10</option></select></div></div>
<div id="sv" class="st"></div><button class="btn bp" onclick="saveConfig()">💾 Save</button>
</div></div>
<div class="tc" id="t-queries"><div id="recent-loading">Loading query log…</div></div>
<div class="tc" id="t-system">
<div class="card"><h2>⚙️ Controls</h2>
<div class="fg"><div class="toggle"><input type="checkbox" id="c-maintenance_mode" {'checked' if config.get('maintenance_mode')=='true' else ''}><label><strong>Maintenance Mode</strong></label></div></div>
<div class="fg"><label>Maintenance Message</label><textarea id="c-maintenance_message" rows="2">{esc(config.get('maintenance_message',''))}</textarea></div>
<div style="display:flex;gap:8px;margin-top:10px;flex-wrap:wrap">
<button class="btn bp" onclick="saveConfig()">💾 Save Config</button>
<button class="btn bg" onclick="rebuildIdx()">🔄 Rebuild RAG</button>
<button class="btn" style="background:#dc2626;color:#fff;border:none;padding:8px 14px;border-radius:8px;font-weight:600;cursor:pointer" onclick="restartSpace()">♻️ Restart Space</button>
</div>
<div id="sys" class="st"></div>
</div>
<div class="card"><h2>🟢 Live Status</h2>
<div id="status-panel" style="font-size:.85rem">Loading status…</div>
</div>
<div class="card"><h2>System Info</h2><table>
<tr><td><strong>Knowledge Files</strong></td><td>{len(glob.glob(os.path.join(KNOWLEDGE_DIR,'*.txt')))} files</td></tr>
<tr><td><strong>OpenAI API</strong></td><td>{'✅ Configured' if os.environ.get('OPENAI_API_KEY') else '❌ Missing'}</td></tr>
<tr><td><strong>Anthropic API</strong></td><td>{'✅ Configured' if os.environ.get('ANTHROPIC_API_KEY') else '❌ Missing'}</td></tr>
<tr><td><strong>PRIMO API</strong></td><td>{'✅ Configured' if os.environ.get('PRIMO_API_KEY') else '❌ Missing'}</td></tr>
<tr><td><strong>Admin Password</strong></td><td>{'✅ Set' if os.environ.get('ADMIN_PASSWORD') else '❌ Missing'}</td></tr>
<tr><td><strong>Cloudflare D1</strong></td><td>Connected via Worker</td></tr>
</table></div></div>
<script>
const W='{WORKER_URL}';
function st(el,t){{document.querySelectorAll('.tab').forEach(e=>e.classList.remove('active'));document.querySelectorAll('.tc').forEach(e=>e.classList.remove('active'));el.classList.add('active');document.getElementById('t-'+t).classList.add('active');}}
// Fetch analytics from Cloudflare D1
fetch(W+'/analytics').then(r=>r.json()).then(d=>{{
const el=document.getElementById('t-analytics');
el.innerHTML=`
<div class="grid">
<div class="stat"><div class="n">${{d.total}}</div><div class="l">Total</div></div>
<div class="stat"><div class="n">${{d.today}}</div><div class="l">Today</div></div>
<div class="stat"><div class="n">${{d.week}}</div><div class="l">This Week</div></div>
<div class="stat"><div class="n">${{(d.avg_time||0).toFixed(1)}}s</div><div class="l">Avg Time</div></div>
<div class="stat"><div class="n">${{d.errors}}</div><div class="l">Errors</div></div>
</div>
<div class="two">
<div class="card"><h2>Tool Usage</h2><table><tr><th>Tool</th><th>Count</th></tr>${{d.tools.map(t=>`<tr><td>${{t.tool_used}}</td><td>${{t.c}}</td></tr>`).join('')}}</table></div>
<div class="card"><h2>Model Usage</h2><table><tr><th>Model</th><th>Count</th></tr>${{d.models.map(m=>`<tr><td>${{m.model}}</td><td>${{m.c}}</td></tr>`).join('')}}</table></div>
</div>
<div class="two">
<div class="card"><h2>Hourly</h2><canvas id="hc"></canvas></div>
<div class="card"><h2>Daily (14d)</h2><canvas id="dc"></canvas></div>
</div>
<div class="card"><h2>Popular (Top 20)</h2><table><tr><th>Question</th><th>Count</th></tr>${{d.popular.map(p=>`<tr><td>${{(p.question||'').substring(0,70)}}</td><td>${{p.c}}</td></tr>`).join('')}}</table></div>`;
if(d.hourly.length)new Chart(document.getElementById('hc'),{{type:'bar',data:{{labels:d.hourly.map(h=>h.hour+':00'),datasets:[{{label:'Q',data:d.hourly.map(h=>h.c),backgroundColor:'#003366'}}]}},options:{{responsive:true,plugins:{{legend:{{display:false}}}}}}}});
if(d.daily.length)new Chart(document.getElementById('dc'),{{type:'line',data:{{labels:d.daily.map(x=>(x.day||'').slice(5)),datasets:[{{label:'Q',data:d.daily.map(x=>x.c),borderColor:'#003366',backgroundColor:'rgba(0,51,102,0.1)',fill:true,tension:.3}}]}},options:{{responsive:true,plugins:{{legend:{{display:false}}}}}}}});
}}).catch(e=>{{document.getElementById('t-analytics').innerHTML='<div class="card" style="color:#dc2626">Failed to load analytics: '+e.message+'<br>Make sure D1 is initialized: <a href="'+W+'/analytics/init" target="_blank">Click here to init DB</a></div>';}});
// Fetch local feedback stats
fetch('/admin/feedback-stats').then(r=>r.json()).then(d=>{{
const host=document.getElementById('t-analytics');
if(!host) return;
const s=(d.summary||[]).reduce((a,x)=>{{a[x.feedback]=x.c;return a;}},{{}});
const card=document.createElement('div');
card.className='card';
card.innerHTML=`<h2>👍 User Feedback</h2>
<div class="grid" style="margin-bottom:10px">
<div class="stat"><div class="n">${{s.up||0}}</div><div class="l">Helpful</div></div>
<div class="stat"><div class="n">${{s.down||0}}</div><div class="l">Not Helpful</div></div>
</div>
<div class="two">
<div><h2 style="margin-bottom:6px">Recent feedback</h2><table><tr><th>Time</th><th>Question</th><th>Feedback</th></tr>${{(d.recent||[]).map(r=>`<tr><td>${{(r.timestamp||'').substring(0,19)}}</td><td>${{(r.question||'').substring(0,60)}}</td><td>${{r.feedback==='up'?'👍':'👎'}}</td></tr>`).join('')}}</table></div>
<div><h2 style="margin-bottom:6px">Top weak questions</h2><table><tr><th>Question</th><th>Count</th></tr>${{(d.weak||[]).map(r=>`<tr><td>${{(r.question||'').substring(0,70)}}</td><td>${{r.c}}</td></tr>`).join('')}}</table></div>
</div>`;
host.appendChild(card);
}}).catch(()=>{{}});
// Fetch recent queries
fetch(W+'/analytics/recent').then(r=>r.json()).then(d=>{{
const el=document.getElementById('t-queries');
el.innerHTML='<div class="card"><h2>Recent (50)</h2><table><tr><th>Time</th><th>Question</th><th>Tool</th><th>Model</th><th>Results</th><th>Status</th></tr>'+(d.results||[]).map(r=>`<tr><td>${{(r.timestamp||'').substring(0,19)}}</td><td title="${{r.question}}">${{(r.question||'').substring(0,55)}}</td><td><span class="badge" style="background:#e0e7ff">${{r.tool_used}}</span></td><td>${{r.model}}</td><td>${{r.result_count}}</td><td class="err">${{r.error||'✓'}}</td></tr>`).join('')+'</table></div>';
}}).catch(e=>{{document.getElementById('t-queries').innerHTML='<div class="card" style="color:#dc2626">'+e.message+'</div>';}});
async function saveConfig(){{
const keys=['welcome_message','bot_personality','custom_instructions','announcement','default_model','max_results','maintenance_message'];
const data={{}};keys.forEach(k=>{{data[k]=document.getElementById('c-'+k).value;}});
data.maintenance_mode=document.getElementById('c-maintenance_mode').checked?'true':'false';
const r=await fetch('/admin/config',{{method:'POST',headers:{{'Content-Type':'application/json'}},body:JSON.stringify(data)}});
const res=await r.json();
const s=document.getElementById('sv')||document.getElementById('sys');
s.className=res.status==='ok'?'st ok':'st er'; s.textContent=res.status==='ok'?'✅ Saved!':'❌ Error';
setTimeout(()=>s.style.display='none',3000);
}}
async function rebuildIdx(){{
const s=document.getElementById('sys');s.className='st ok';s.textContent='🔄 Rebuilding...';s.style.display='block';
const r=await fetch('/rebuild',{{method:'POST'}});const res=await r.json();
s.textContent=res.status==='ok'?'✅ Rebuilt: '+res.chunks+' chunks from '+res.documents+' files':'❌ '+(res.error||JSON.stringify(res));
s.className=res.status==='ok'?'st ok':'st er';
loadStatus(); // refresh status after rebuild
}}
async function restartSpace(){{
if(!confirm('Restart the HF Space? The service will be unavailable for ~30-60 seconds.')) return;
const s=document.getElementById('sys');s.className='st ok';s.textContent='♻️ Restarting...';s.style.display='block';
try {{
await fetch('/restart',{{method:'POST'}});
s.textContent='♻️ Space restarting — please wait 30-60 seconds then refresh this page.';
// Poll until back online
let attempts=0;
const poll=setInterval(async()=>{{
attempts++;
try{{
const r=await fetch('/status');
if(r.ok){{clearInterval(poll);s.textContent='✅ Space is back online!';s.className='st ok';loadStatus();}}
}}catch(e){{
s.textContent=`♻️ Still restarting... (${{attempts*5}}s elapsed)`;
}}
if(attempts>24){{clearInterval(poll);s.textContent='⚠️ Restart taking longer than expected. Try refreshing the page.';}}
}},5000);
}}catch(e){{s.textContent='❌ '+e.message;s.className='st er';}}
}}
async function loadStatus(){{
const el=document.getElementById('status-panel');
if(!el) return;
try{{
const r=await fetch('/status');
if(!r.ok){{el.innerHTML='<span style="color:#dc2626">❌ HF Space not responding</span>';return;}}
const d=await r.json();
const uptimeMin=Math.round((d.uptime_secs||0)/60);
const memColor=d.mem_pct>85?'#dc2626':d.mem_pct>70?'#d97706':'#16a34a';
el.innerHTML=`
<table style="width:100%">
<tr><td><strong>Space Status</strong></td><td><span style="color:#16a34a;font-weight:700">✅ Running</span></td></tr>
<tr><td><strong>Vectorstore</strong></td><td>${{d.vectorstore==='ready'?'✅ Ready ('+d.chunks+' chunks, '+d.kb_files+' files)':'❌ Not initialized'}}</td></tr>
<tr><td><strong>OpenAI API</strong></td><td>${{d.openai?'✅ Configured':'❌ Missing'}}</td></tr>
<tr><td><strong>Anthropic API</strong></td><td>${{d.anthropic?'✅ Configured':'❌ Missing'}}</td></tr>
<tr><td><strong>PRIMO API</strong></td><td>${{d.primo?'✅ Configured':'❌ Missing'}}</td></tr>
<tr><td><strong>Memory Usage</strong></td><td><span style="color:${{memColor}};font-weight:700">${{d.mem_pct}}%</span> (${{d.mem_used_mb}}MB / ${{d.mem_total_mb}}MB)</td></tr>
<tr><td><strong>Uptime</strong></td><td>${{uptimeMin < 60 ? uptimeMin+'m' : Math.floor(uptimeMin/60)+'h '+uptimeMin%60+'m'}}</td></tr>
</table>
<div style="margin-top:8px;font-size:.74rem;color:#6b7280">Last checked: ${{new Date().toLocaleTimeString()}}</div>`;
}}catch(e){{
el.innerHTML='<span style="color:#dc2626">❌ Cannot reach HF Space: '+e.message+'</span>';
}}
}}
// Load status on page load and every 30 seconds
loadStatus();
setInterval(loadStatus, 30000);
</script>
</body></html>"""
# ---- Restart Space endpoint ----
@app.post("/restart")
async def restart_space(request: Request):
if not verify_session(request):
raise HTTPException(status_code=401, detail="Not authenticated")
import threading
def do_restart():
import time as t
t.sleep(1)
os.kill(os.getpid(), 9) # Force restart — HF Space will auto-restart
threading.Thread(target=do_restart, daemon=True).start()
return {"status": "restarting", "message": "Space is restarting. Please wait 30-60 seconds."}
# ---- Status endpoint (no auth needed — for frontend health checks) ----
@app.get("/status")
async def get_status():
# Read container-level memory from /sys/fs/cgroup (accurate for HF Spaces)
# psutil reads the host machine memory which is misleading on shared containers
try:
# cgroup v2 (HF Spaces uses this)
with open('/sys/fs/cgroup/memory.current') as f:
mem_used_bytes = int(f.read().strip())
with open('/sys/fs/cgroup/memory.max') as f:
val = f.read().strip()
mem_total_bytes = int(val) if val != 'max' else 16 * 1024**3 # 16GB default
mem_used_mb = round(mem_used_bytes / 1024 / 1024)
mem_total_mb = round(mem_total_bytes / 1024 / 1024)
mem_pct = round(mem_used_bytes / mem_total_bytes * 100)
except:
# Fallback: try cgroup v1
try:
with open('/sys/fs/cgroup/memory/memory.usage_in_bytes') as f:
mem_used_bytes = int(f.read().strip())
with open('/sys/fs/cgroup/memory/memory.limit_in_bytes') as f:
mem_total_bytes = int(f.read().strip())
mem_used_mb = round(mem_used_bytes / 1024 / 1024)
mem_total_mb = round(mem_total_bytes / 1024 / 1024)
mem_pct = round(mem_used_bytes / mem_total_bytes * 100)
except:
mem_used_mb = mem_total_mb = mem_pct = 0
kb_files = glob.glob(os.path.join(KNOWLEDGE_DIR, "*.txt"))
chunks = vectorstore.index.ntotal if vectorstore else 0
return {
"status": "ok",
"vectorstore": "ready" if vectorstore else "not_initialized",
"chunks": chunks,
"kb_files": len(kb_files),
"openai": bool(os.environ.get("OPENAI_API_KEY")),
"anthropic": bool(os.environ.get("ANTHROPIC_API_KEY")),
"primo": bool(os.environ.get("PRIMO_API_KEY")),
"mem_used_mb": mem_used_mb,
"mem_total_mb": mem_total_mb,
"mem_pct": mem_pct,
"uptime_secs": round(time.time() - START_TIME),
}
# ---- Admin stats from D1 (kept for API compatibility) ----
@app.get("/admin/stats")
async def admin_stats(request: Request):
if not verify_session(request):
raise HTTPException(status_code=401, detail="Not authenticated")
# Stats now come from Cloudflare D1 via /analytics endpoint
return {"message": "Use Cloudflare Worker /analytics endpoint for D1 data"}
conn = sqlite3.connect(DB_PATH)
conn.row_factory = sqlite3.Row
total = conn.execute("SELECT COUNT(*) as c FROM queries").fetchone()["c"]
today = conn.execute("SELECT COUNT(*) as c FROM queries WHERE timestamp >= date('now')").fetchone()["c"]
tools = {r["tool_used"]: r["c"] for r in conn.execute("SELECT tool_used, COUNT(*) as c FROM queries GROUP BY tool_used").fetchall()}
models = {r["model"]: r["c"] for r in conn.execute("SELECT model, COUNT(*) as c FROM queries GROUP BY model").fetchall()}
conn.close()
return {"total": total, "today": today, "tools": tools, "models": models}
# ---- Admin Config GET/POST (protected) ----
@app.get("/admin/config")
async def admin_get_config(request: Request):
if not verify_session(request):
raise HTTPException(status_code=401, detail="Not authenticated")
conn = sqlite3.connect(DB_PATH)
conn.row_factory = sqlite3.Row
rows = conn.execute("SELECT key, value, updated_at FROM bot_config").fetchall()
conn.close()
return {r["key"]: {"value": r["value"], "updated_at": r["updated_at"]} for r in rows}
@app.post("/admin/config")
async def admin_set_config(request: Request):
if not verify_session(request):
raise HTTPException(status_code=401, detail="Not authenticated")
data = await request.json()
settings = data.get("settings", data) if isinstance(data, dict) else data
for key, value in settings.items():
set_config(key, str(value))
return {"status": "ok", "updated": list(settings.keys())}
class FeedbackRequest(BaseModel):
question: str = ""
feedback: str
answer_excerpt: str = ""
intent: str = ""
@app.post("/feedback")
async def feedback_submit(req: FeedbackRequest):
try:
conn = sqlite3.connect(DB_PATH)
conn.execute(
"INSERT INTO feedback (timestamp, question, feedback, answer_excerpt, intent) VALUES (?,?,?,?,?)",
(datetime.utcnow().isoformat(), req.question or "", req.feedback or "", (req.answer_excerpt or "")[:500], req.intent or "")
)
conn.commit()
conn.close()
return {"status": "ok"}
except Exception as e:
return {"status": "error", "error": str(e)}
@app.get("/admin/feedback-stats")
async def admin_feedback_stats(request: Request):
if not verify_session(request):
raise HTTPException(status_code=401, detail="Not authenticated")
conn = sqlite3.connect(DB_PATH)
conn.row_factory = sqlite3.Row
summary = [dict(r) for r in conn.execute("SELECT feedback, COUNT(*) as c FROM feedback GROUP BY feedback ORDER BY c DESC").fetchall()]
recent = [dict(r) for r in conn.execute("SELECT timestamp, question, feedback, intent FROM feedback ORDER BY id DESC LIMIT 30").fetchall()]
weak = [dict(r) for r in conn.execute("SELECT question, COUNT(*) as c FROM feedback WHERE feedback='down' AND question != '' GROUP BY question ORDER BY c DESC LIMIT 10").fetchall()]
conn.close()
return {"summary": summary, "recent": recent, "weak": weak}
# ---- Public config (for frontend) ----
@app.get("/config")
async def public_config():
return {
"welcome_message": get_config("welcome_message"),
"bot_personality": get_config("bot_personality"),
"custom_instructions": get_config("custom_instructions"),
"announcement": get_config("announcement"),
"default_model": get_config("default_model", "gpt"),
"maintenance_mode": get_config("maintenance_mode", "false"),
"maintenance_message": get_config("maintenance_message"),
"max_results": get_config("max_results", "5"),
}
# ---- Clear logs ----
@app.post("/admin/clear-logs")
async def clear_logs():
conn = sqlite3.connect(DB_PATH)
conn.execute("DELETE FROM queries")
conn.commit()
conn.close()
return {"status": "ok", "message": "All logs cleared"}
|