Hybrid Document AI — KIE field classifier (v1)
scikit-learn field classifier for receipt KIE (Tier-1 of a hybrid OCR+VLM pipeline).
Features: OCR confidence, layout position, keyword proximity, money/date patterns.
End-to-end macro-F1 = 0.865 (40 receipts, real RapidOCR). See github.com/NguyenDucAnforwork/hybrid-document-ai