Spaces:
Paused
Paused
File size: 23,733 Bytes
0d3f7cc | 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 | """
Hermes Tools — Unified compliance pipeline and engine access.
This module provides:
- `CompliancePipeline`: Thread-safe orchestrator for PDF parsing, HS classification,
and sanctions screening in a single call.
- Re-exports from `base` for backward compatibility.
Usage:
from hermes.tools import CompliancePipeline, get_pipeline
pipeline = get_pipeline()
review_item = pipeline.process_incoming_document(
file_path="invoice.pdf",
context={"shipper": "Acme Corp", "consignee": "Global Imports LLC"},
)
"""
from __future__ import annotations
import logging
import tempfile
import threading
from pathlib import Path
from typing import Annotated, Any, Final, Optional
from hermes.tools.base.registry import tool_registry
from hermes.tools.base.tool import BaseTool
logger: Final = logging.getLogger(__name__)
# ── Domain Exceptions ─────────────────────────────────────────────────
class PipelineError(Exception):
"""Base exception for pipeline execution errors."""
class PipelineValidationError(PipelineError):
"""Raised when pipeline input validation fails.
Attributes:
field: The field that failed validation.
reason: Description of the validation failure.
"""
def __init__(self, field: str, reason: str) -> None:
self.field = field
self.reason = reason
super().__init__(f"Validation failed for '{field}': {reason}")
class PipelinePDFParsingError(PipelineError):
"""Raised when PDF parsing fails during pipeline execution.
Attributes:
file_path: Path to the PDF that failed parsing.
original_error: The underlying parsing exception.
"""
def __init__(self, file_path: str, original_error: Exception) -> None:
self.file_path = file_path
self.original_error = original_error
super().__init__(f"PDF parsing failed for {file_path}: {original_error}")
class PipelineClassificationError(PipelineError):
"""Raised when HS code classification fails.
Attributes:
description: The product description that failed classification.
original_error: The underlying classification exception.
"""
def __init__(self, description: str, original_error: Exception) -> None:
self.description = description
self.original_error = original_error
super().__init__(
f"HS classification failed for '{description[:50]}...': {original_error}"
)
class PipelineSanctionsError(PipelineError):
"""Raised when sanctions screening fails.
Attributes:
party_name: The party name that failed screening.
original_error: The underlying screening exception.
"""
def __init__(self, party_name: str, original_error: Exception) -> None:
self.party_name = party_name
self.original_error = original_error
super().__init__(
f"Sanctions screening failed for '{party_name}': {original_error}"
)
# ── Constants ─────────────────────────────────────────────────────────
MAX_TEMP_FILE_SIZE: Final[int] = 50 * 1024 * 1024 # 50 MB
# ── Compliance Pipeline ───────────────────────────────────────────────
class CompliancePipeline:
"""Thread-safe pipeline orchestrating PDF parsing, HS classification, and sanctions screening.
Provides a single method `process_incoming_document` that:
1. Parses the PDF to extract document fields
2. Classifies extracted items to HS codes
3. Screens all parties against sanctions lists
4. Assembles a ReviewItem for human approval
All errors are wrapped in pipeline-specific exceptions for consistent
error handling upstream.
Attributes:
_parser: PDF parser singleton.
_hs_engine: HS classification engine singleton.
_sanctions_engine: Sanctions screening engine singleton.
"""
def __init__(self) -> None:
"""Initialize the pipeline with engine singletons."""
from hermes.tools.pdf_parser import get_parser
from hermes.tools.hs_classifier import get_engine as get_hs_engine
from hermes.tools.sanctions import get_engine as get_sanctions_engine
self._parser = get_parser()
self._hs_engine = get_hs_engine()
self._sanctions_engine = get_sanctions_engine()
self._lock: threading.Lock = threading.Lock()
def process_incoming_document(
self,
file_path: str,
context: Optional[dict[str, Any]] = None,
) -> dict[str, Any]:
"""Process a trade document through the full compliance pipeline.
Orchestrates PDF parsing → HS classification → sanctions screening
and returns a ReviewItem dict ready for human review.
Args:
file_path: Path to the PDF document to process.
context: Optional context dict with overrides:
- shipper (str): Override shipper name.
- consignee (str): Override consignee name.
- country_origin (str): Override country of origin.
- country_destination (str): Override destination country.
- invoice_number (str): Override invoice number.
Returns:
Dict representing a ReviewItem with all extracted and classified data.
Raises:
FileNotFoundError: If the file does not exist.
PipelineValidationError: If input validation fails.
PipelinePDFParsingError: If PDF parsing fails.
PipelineClassificationError: If HS classification fails.
PipelineSanctionsError: If sanctions screening fails.
"""
ctx: dict[str, Any] = context or {}
path: Path = Path(file_path)
if not path.exists():
raise FileNotFoundError(f"Document not found: {file_path}")
if not path.suffix.lower() == ".pdf":
raise PipelineValidationError(
"file_path",
f"Expected PDF file, got {path.suffix or 'no extension'}",
)
# Step 1: Parse PDF
parsed_doc = self._parse_pdf(file_path)
# Step 2: Extract and classify items
classifications: list[dict[str, Any]] = self._classify_items(
parsed_doc, ctx
)
# Step 3: Screen parties
sanctions_results: dict[str, Any] = self._screen_parties(parsed_doc, ctx)
# Step 4: Assemble ReviewItem
review_item: dict[str, Any] = self._assemble_review_item(
file_path=file_path,
parsed_doc=parsed_doc,
classifications=classifications,
sanctions_results=sanctions_results,
context=ctx,
)
logger.info(
"Pipeline processed %s → review item %s",
file_path,
review_item.get("item_id", "pending"),
)
return review_item
def _parse_pdf(self, file_path: str) -> Any:
"""Parse the PDF and extract document fields.
Args:
file_path: Path to the PDF file.
Returns:
ParsedDocument with extracted fields.
Raises:
PipelinePDFParsingError: If parsing fails.
"""
try:
from hermes.tools.pdf_parser import (
PDFSizeLimitExceededError,
PDFPageLimitExceededError,
PDFNotValidError,
PDFEmptyError,
)
return self._parser.parse(file_path)
except (PDFSizeLimitExceededError, PDFPageLimitExceededError) as exc:
raise PipelinePDFParsingError(file_path, exc) from exc
except (PDFNotValidError, PDFEmptyError) as exc:
raise PipelinePDFParsingError(file_path, exc) from exc
except FileNotFoundError:
raise
except Exception as exc:
raise PipelinePDFParsingError(file_path, exc) from exc
def _classify_items(self, parsed_doc: Any, ctx: dict[str, Any]) -> list[dict[str, Any]]:
"""Classify extracted items to HS codes.
Args:
parsed_doc: ParsedDocument from PDF parsing.
ctx: Context dict with optional overrides.
Returns:
List of classification result dicts.
"""
from hermes.tools.hs_classifier import HSClassificationHallucinationError
from hermes.tools.pdf_parser import InvoiceItem
classifications: list[dict[str, Any]] = []
items: list[InvoiceItem] = getattr(parsed_doc, "items", [])
if not items:
# If no items extracted, classify based on document description
description: str = self._build_description_from_doc(parsed_doc, ctx)
if description:
cls_result = self._classify_single(description, ctx)
if cls_result:
classifications.append(cls_result)
return classifications
country_origin: str = ctx.get(
"country_origin",
getattr(
parsed_doc.country_origin, "value", ""
) if parsed_doc.country_origin else "",
)
country_dest: str = ctx.get(
"country_destination",
getattr(
parsed_doc.country_destination, "value", ""
) if parsed_doc.country_destination else "",
)
for item in items:
if not item.description:
continue
try:
response = self._hs_engine.classify(
description=item.description,
country_origin=country_origin,
country_destination=country_dest,
use_llm=False,
target_digits=6,
)
alt_list: list[dict[str, Any]] = []
for alt in response.alternatives:
alt_list.append({
"hs_code": alt.hs_code,
"description": alt.description,
"confidence": round(alt.confidence, 3),
})
classifications.append({
"description": item.description,
"hs_code": response.primary.hs_code,
"hs_description": response.primary.description,
"confidence": round(response.primary.confidence, 3),
"source": response.primary.source,
"needs_human_review": response.primary.needs_human_review,
"alternatives": alt_list,
"quantity": item.quantity,
"unit_price": item.unit_price,
"total_price": item.total_price,
})
except HSClassificationHallucinationError as exc:
logger.warning(
"Classification hallucination detected for '%s': %s",
item.description[:50],
exc,
)
classifications.append({
"description": item.description,
"hs_code": "999999",
"hs_description": "Classification failed - hallucination detected",
"confidence": 0.0,
"source": "error",
"needs_human_review": True,
"alternatives": [],
"quantity": item.quantity,
"unit_price": item.unit_price,
"total_price": item.total_price,
})
except Exception as exc:
logger.warning(
"Classification failed for '%s': %s",
item.description[:50],
exc,
)
classifications.append({
"description": item.description,
"hs_code": "",
"hs_description": f"Classification error: {exc}",
"confidence": 0.0,
"source": "error",
"needs_human_review": True,
"alternatives": [],
"quantity": item.quantity,
"unit_price": item.unit_price,
"total_price": item.total_price,
})
return classifications
def _classify_single(
self, description: str, ctx: dict[str, Any]
) -> Optional[dict[str, Any]]:
"""Classify a single description.
Args:
description: Product description.
ctx: Context dict with optional overrides.
Returns:
Classification result dict, or None if classification fails.
"""
try:
country_origin: str = ctx.get("country_origin", "")
country_dest: str = ctx.get("country_destination", "")
response = self._hs_engine.classify(
description=description,
country_origin=country_origin,
country_destination=country_dest,
use_llm=False,
target_digits=6,
)
alt_list: list[dict[str, Any]] = []
for alt in response.alternatives:
alt_list.append({
"hs_code": alt.hs_code,
"description": alt.description,
"confidence": round(alt.confidence, 3),
})
return {
"description": description,
"hs_code": response.primary.hs_code,
"hs_description": response.primary.description,
"confidence": round(response.primary.confidence, 3),
"source": response.primary.source,
"needs_human_review": response.primary.needs_human_review,
"alternatives": alt_list,
"quantity": 0,
"unit_price": 0,
"total_price": 0,
}
except Exception as exc:
logger.warning("Single classification failed: %s", exc)
return None
def _build_description_from_doc(
self, parsed_doc: Any, ctx: dict[str, Any]
) -> str:
"""Build a product description from parsed document fields.
Args:
parsed_doc: ParsedDocument.
ctx: Context dict.
Returns:
Description string for classification.
"""
parts: list[str] = []
if parsed_doc.document_type and parsed_doc.document_type != "unknown":
parts.append(parsed_doc.document_type.replace("_", " ").title())
if ctx.get("description"):
parts.append(ctx["description"])
if ctx.get("product_description"):
parts.append(ctx["product_description"])
return " ".join(parts) if parts else ""
def _screen_parties(
self, parsed_doc: Any, ctx: dict[str, Any]
) -> dict[str, Any]:
"""Screen all parties against sanctions lists.
Args:
parsed_doc: ParsedDocument with party information.
ctx: Context dict with optional overrides.
Returns:
Dict mapping party names to their screening results.
"""
from hermes.tools.sanctions import ScreeningResult
parties: dict[str, str] = {}
shipper: str = ctx.get("shipper", "")
if not shipper and parsed_doc.shipper:
shipper = getattr(parsed_doc.shipper, "name", "")
if shipper:
parties["shipper"] = shipper
consignee: str = ctx.get("consignee", "")
if not consignee and parsed_doc.consignee:
consignee = getattr(parsed_doc.consignee, "name", "")
if consignee:
parties["consignee"] = consignee
notify_party: str = ctx.get("notify_party", "")
if not notify_party and parsed_doc.notify_party:
notify_party = getattr(parsed_doc.notify_party, "name", "")
if notify_party:
parties["notify_party"] = notify_party
results: dict[str, Any] = {}
for role, name in parties.items():
try:
result: ScreeningResult = self._sanctions_engine.screen(name)
matches_data: list[dict[str, Any]] = []
for match in result.matches:
matches_data.append({
"entity_name": match.entity.name,
"confidence": round(match.confidence, 3),
"match_type": match.match_type,
"matched_name": match.matched_name,
"source": match.source,
})
results[role] = {
"name": name,
"risk_level": result.risk_level,
"has_matches": result.has_matches,
"matches": matches_data,
"sources_checked": result.sources_checked,
}
except Exception as exc:
logger.warning(
"Sanctions screening failed for %s '%s': %s",
role,
name[:50],
exc,
)
results[role] = {
"name": name,
"risk_level": "unknown",
"has_matches": False,
"matches": [],
"sources_checked": [],
"error": str(exc),
}
return results
def _assemble_review_item(
self,
file_path: str,
parsed_doc: Any,
classifications: list[dict[str, Any]],
sanctions_results: dict[str, Any],
context: dict[str, Any],
) -> dict[str, Any]:
"""Assemble a ReviewItem dict from pipeline results.
Args:
file_path: Original PDF file path.
parsed_doc: ParsedDocument from PDF parsing.
classifications: List of HS classification results.
sanctions_results: Dict of sanctions screening results.
context: Context dict with overrides.
Returns:
Dict representing a ReviewItem.
"""
invoice_number: str = context.get("invoice_number", "")
if not invoice_number and parsed_doc.invoice_number:
invoice_number = getattr(parsed_doc.invoice_number, "value", "")
invoice_date: str = context.get("invoice_date", "")
if not invoice_date and parsed_doc.invoice_date:
invoice_date = getattr(parsed_doc.invoice_date, "value", "")
total_amount: str = context.get("total_amount", "")
if not total_amount and parsed_doc.total_amount:
total_amount = getattr(parsed_doc.total_amount, "value", "")
shipper: str = context.get("shipper", "")
if not shipper and parsed_doc.shipper:
shipper = getattr(parsed_doc.shipper, "name", "")
consignee: str = context.get("consignee", "")
if not consignee and parsed_doc.consignee:
consignee = getattr(parsed_doc.consignee, "name", "")
country_origin: str = context.get("country_origin", "")
if not country_origin and parsed_doc.country_origin:
country_origin = getattr(parsed_doc.country_origin, "value", "")
country_dest: str = context.get("country_destination", "")
if not country_dest and parsed_doc.country_destination:
country_dest = getattr(parsed_doc.country_destination, "value", "")
primary_hs: str = ""
hs_description: str = ""
hs_confidence: float = 0.0
hs_alternatives: list[dict[str, Any]] = []
needs_human_review: bool = False
if classifications:
best: dict[str, Any] = max(
classifications, key=lambda c: c.get("confidence", 0)
)
primary_hs = best.get("hs_code", "")
hs_description = best.get("hs_description", "")
hs_confidence = best.get("confidence", 0.0)
needs_human_review = best.get("needs_human_review", False)
hs_alternatives = best.get("alternatives", [])
sanctions_risk: str = "clear"
all_sanctions_matches: list[dict[str, Any]] = []
for role, result in sanctions_results.items():
level: str = result.get("risk_level", "clear")
if level in ("high", "blocked"):
sanctions_risk = level
for match in result.get("matches", []):
all_sanctions_matches.append({
"role": role,
**match,
})
if needs_human_review or sanctions_risk in ("high", "blocked"):
priority: int = 2 if sanctions_risk == "blocked" else 1
else:
priority = 0
return {
"item_id": "",
"document_path": file_path,
"document_type": parsed_doc.document_type or "unknown",
"invoice_number": invoice_number,
"invoice_date": invoice_date,
"total_amount": total_amount,
"shipper": shipper,
"consignee": consignee,
"country_origin": country_origin,
"country_destination": country_dest,
"hs_code_suggested": primary_hs,
"hs_code_description": hs_description,
"hs_code_confidence": hs_confidence,
"hs_code_alternatives": hs_alternatives,
"sanctions_risk_level": sanctions_risk,
"sanctions_matches": all_sanctions_matches,
"status": "pending",
"assigned_to": "",
"priority": priority,
"comments": [],
"actions": [],
"final_hs_code": "",
"reviewer_notes": "",
}
# ── Singleton ─────────────────────────────────────────────────────────
_pipeline: Optional[CompliancePipeline] = None
_pipeline_lock: threading.Lock = threading.Lock()
def get_pipeline() -> CompliancePipeline:
"""Get or create the global compliance pipeline (thread-safe).
Returns:
The singleton CompliancePipeline instance.
"""
global _pipeline
if _pipeline is None:
with _pipeline_lock:
if _pipeline is None:
_pipeline = CompliancePipeline()
return _pipeline
# ── Backward Compatibility ────────────────────────────────────────────
__all__ = [
"BaseTool",
"tool_registry",
"CompliancePipeline",
"PipelineError",
"PipelineValidationError",
"PipelinePDFParsingError",
"PipelineClassificationError",
"PipelineSanctionsError",
"get_pipeline",
]
|