Overview
VelaAI was designed to automate the most tedious and error-prone part of small-to-medium business accounting — manually entering receipts. Staff sends receipts via WhatsApp, the system reads them automatically, suggests account codes, and the accountant approves.
System Architecture
A multi-tenant structure:
- Admin layer: Platform owner managing the system (manages companies)
- Company layer: Customer companies receiving service
- Accountant layer: Accountants serving each company
- Staff layer: End users sending receipts
Key Features
- WhatsApp integration: Receipt intake from staff and accountant approval flow via Meta Graph API
- OCR pipeline: Automatic extraction of receipt items (amount, VAT, date, vendor) using Mestral OCR
- Self-learning account matching: A system that matches receipt descriptions to sub-account codes using Sentence-Transformers embeddings; learns from every approval and becomes more accurate over time
- Flutter desktop panel: Desktop application where accountants review and process pending receipts
- ASP.NET admin platform: Web panel where the platform owner manages all companies, accountants, and system health
Technical Details
The backend is split between ASP.NET and FastAPI: ASP.NET handles enterprise user/company/permission management, while FastAPI handles OCR and embedding inferences. SQL Server is used for enterprise data, and a vector store for embedding matching.