iil-ingest

Reusable Document Ingestion Package for the IIL Platform — ADR-170

Pattern: iil-enrichment (ADR-169) — Pure Python Core, Protocol-based, optional Django integration

Installation

pip install iil-ingest[pdf]        # PDF only
pip install iil-ingest[all]        # PDF + Excel + DOCX
pip install iil-ingest[all,django] # + Django mixins

Quick Start

from ingest import IngestPipeline, ProfileClassifier
from ingest.extractors.pdf import PDFExtractor
from ingest.profiles.german_hr import GERMAN_HR_PROFILES

pipeline = IngestPipeline(
    extractors=[PDFExtractor()],
    classifier=ProfileClassifier(GERMAN_HR_PROFILES),
)

with open("rechnung.pdf", "rb") as f:
    doc = pipeline.run(f.read(), filename="rechnung.pdf")

print(doc.doc_type)    # "RECHNUNG"
print(doc.confidence)  # "HIGH"
print(doc.score)       # 42.0

Custom Profile

from dataclasses import dataclass
from ingest import IngestPipeline, ProfileClassifier
from ingest.extractors.pdf import PDFExtractor

@dataclass
class SdsProfile:
    name = "SDS"
    patterns = [
        (r"sicherheitsdatenblatt", 10),
        (r"gefahrenhinweis|h\d{3}", 8),
        (r"reach", 6),
    ]
    min_score = 10

pipeline = IngestPipeline(
    extractors=[PDFExtractor()],
    classifier=ProfileClassifier([SdsProfile()]),
)

Architecture

Upload -> iil-ingest (detect + extract + classify)
       -> iil-enrichment (optional enrichment)
       -> Django DB / Paperless (archive)

Supported Formats

Pipeline Flow

bytes + filename
  -> detect_mime()          # magic bytes + extension
  -> extractor.extract()    # text, tables, metadata
  -> classifier.classify()  # doc_type, confidence, score
  -> IngestedDocument

References

• ADR-170: platform/docs/adr/ADR-170-iil-ingest-document-ingestion-package.md
• Pattern: iil-enrichment (ADR-169)
• Migrated from: dms-hub/apps/benefits/classifier.py, dms-hub/apps/accounting/extractor.py