Documents and primary records: Prodimex and Sotex
«Продимекс» · Sotex
- Process
- 9 screens
- an end-to-end expense report
Task: remove the manual entry of primary records and the manual preparation of legal documents — from a photo or a scan straight to a finished document.
How it works
- 01Scanned documentPDF and photos of varying quality, with no text layer
- 02Photo of a receiptTaken on a phone while travelling
- 03Extraction and checkingPyMuPDF and Tesseract on the printed layer, the model on everything else
- 04Structured dataFiscal details ФД, ФП, ФН and ККТ, sums, line items
- 05Validation by rulesCompany limits and forbidden positions; anything doubtful goes to a person
- 06OutputLegal documents as docx — board minutes, claims and lawsuits — and a nine-screen expense report tied to meetings from the ELMA export
Particulars and difficulties
The quality of the source
Crumpled receipts, glare and scans with no text layer — OCR alone does not cope
Fiscal details to the digit
The report reconciles with the fiscal data on ФД, ФП, ФН and ККТ: an error breaks the whole filing
Regulation instead of code
Limits and forbidden positions differ per company — they are configured, not programmed
Legal precision
A protocol or a lawsuit cannot be approximate: the structure and wording are fixed and generation follows them
Results
- 9
- screens of the end-to-end process
- 4
- document types for the legal team
- 2
- layers of checking: rules and limits
Primary records enter themselves
An employee photographs a receipt instead of retyping it into a form
A lawyer edits rather than writes from scratch
A protocol, a claim and a lawsuit are assembled from the case data into a finished docx
Stack
FastAPI · PyMuPDF · Tesseract · мультимодальные LLM · python-docx · JWT · pytest