AI Invoicing SaaS — Production Hardening

Context

An AI invoicing SaaS originally built on Replit needed to move from working prototype to software that could be trusted in production. The engagement was a hardening pass over the existing codebase rather than new feature work: security, structure, validation, and tests.

The Hard Part

AI-generated apps ship working features fast but arrive with no authorization boundaries, no input validation, and no tests — the rebuild was not about features, it was about everything underneath them.

What I Built

• Security audit of the Replit-built codebase
• Modularization of source files that had grown past 1,000 lines
• Strict TypeScript throughout, with Zod schemas validating input
• Rate limiting and CSRF protection
• Jest and Playwright test suite over the hardened codebase

Tech Stack