The company processed over 100 documents daily — invoices, acts of completed works, payment requests. Everything was done manually: documents arrived via email, were uploaded to Google Drive, printed, and approved by accountants and managers. Because of this:

• • The average time to process one document was 12–20 minutes.
• • Frequent instances of file duplication or loss.
• • Human errors during completion.
• • Delays in processing payments and submitting reports.

• We proposed the creation of a digital twin of the company's document management system using:

• • AI module for automatic document reading (OCR + NLP)
• • Intelligent approval routing (based on document type, amount, and department)
• • Automatic archive generation with filters and search
• • Integration with CRM and accounting system (1C)

• • Analysis of the structure of existing documents (3 working days)
• • Development of a custom AI model based on GPT + LangChain (5 days)
• • Testing on internal documents — classification accuracy: 96%
• • Integration with the internal file exchange system and email (Gmail API)

• • Average time for processing one document decreased to 1–2 minutes
• • 0% document loss — everything is archived in a unified format
• • Up to 80% of documents are approved automatically without human involvement
• • Total time savings in the first month: over 70 hours
• • Filling errors are minimized