viparse
Python library
v0.1.23 · MIT
Vietnamese documents your RAG stack can actually read. It detects pre-Unicode TCVN3, VNI and VISCII text, converts it to Unicode NFC, and loads DOCX, XLSX, PDF, scanned PDF and legacy .doc through one call.
0.982 diacritic accuracy across 96 hand-transcribed government documents, against 0.019 for the same reader with conversion switched off. Same files, same command, one flag apart.
B¸o c¸o tµi chÝnh quý II n¨m 2026
Báo cáo tài chính quý II năm 2026