Full-Stack + Applied AI · Full-stack engineering project (2026)
Vinayaga Timbers ERP
Production ERP for a four-plant export packaging manufacturer, with Tesseract OCR invoice scanning and bilingual voice input
Harsha's ERP reads photographed invoices with Tesseract OCR and takes shop-floor numbers by voice in Tamil or English. 34 tables, 4 plants, 47 people. I only speak binary.
- 34
- tables
- 4
- plants
- 47
- staff users
Problem
Four plants and 47 staff ran orders, inventory, production, compliance, dispatch, payroll and receivables without a unified system.
What Harsha built
34-table production ERP; Tesseract OCR invoice scanning extracting supplier, line items and tax from photographed bills into a human-review workflow; bilingual (Tamil/English) voice numeric input for shop-floor entry; server-side RBAC and 2FA.
Outcome
Shipped to 4 plants and 47 staff; found and fixed two race conditions (concurrent requests let 56 guesses past a 5-attempt lockout) with row locks and PostgreSQL advisory locks, verified with 20-thread tests.
Stack
- FastAPI
- PostgreSQL
- React
- TypeScript
- Tesseract OCR
- RBAC
- 2FA