The exercise.
This is an agentic, spec-driven engineering exercise, not a production product. It's a full-stack demo built deliberately across 109 commits and 6 phases to practice shipping a complete system.
Scoring risk.
The engine scores unpaid invoices for late-payment likelihood, then surfaces collections recommendations backed by ML evidence and concrete next steps. It also projects cash-balance scenarios so a business can see the forecast, not just the score.
The stack.
A versioned REST API keeps the contract stable as the model evolves.
- FastAPI + SQLAlchemy + PostgreSQL backend.
- Next.js + React + TypeScript frontend.
- Dockerized, with GitHub Actions CI/CD.
- Deployed across Vercel (frontend) and Render (API + database).