Hello
nonso.codes
← All work

MoneyCheck

A pre-spend decision tool: TypeScript calculates whether a spending decision looks manageable based on income; AI chat explains results and does not compute the scores.

Public demo · AI-assisted

Summary

MoneyCheck evaluates a planned purchase with structured metrics before spending. Effective income is monthly earnings plus free or extra money (for example a gift). Deterministic TypeScript calculations produce the scores; AI chat is for educational explanations only.

My role and contribution

I set the product direction and guided an AI-assisted build, and I am the sole GitHub committer. I made sure onboarding asks for a monthly amount, checked the AI chat by hand, and tested the app informally with a couple of people.

Status

Public demo on Vercel (money-check-theta.vercel.app). The demo reflects the initial build and has not had major updates since.

Technology

Next.js, React, TypeScript, Supabase, PostgreSQL, OpenAI SDK, Vercel — as present in the repository and live deployment.

Problem

A purchase price alone rarely shows whether a decision is manageable. MoneyCheck is aimed at students and early-career users who want a quick, structured look at a purchase before they spend.

User workflow

  • If not logged in, middleware sends the user to login.
  • If onboarding is incomplete, the account is not treated as ready.
  • Running a money check saves a record with an id; the result page loads that id for the signed-in user.
  • AI chat can explain results; it does not compute risk, regret, or scores.

What I personally built

  • Guided the AI-assisted implementation (Cursor for building, ChatGPT for planning, Google Stitch for UI).
  • Product checks: monthly-amount onboarding prompt; manual AI chat verification.
  • Informal testing with a couple of people.

The core modules were generated under that guidance, and I'm continuing to study the authentication, calculation and result-loading paths in more depth.

Architecture or data flow

Auth-gated routes and a calc-versus-AI split exist in the codebase: TypeScript calculations produce structured outputs; AI explains those outputs. The repository includes migrations and Row-Level Security policies, though production user-isolation testing has not been documented yet.

Engineering decisions

Keeping deterministic calculations separate from AI explanation is visible in the code and is the most defensible technical pattern to highlight. Scores come from TypeScript, not from the model.

AI-assisted development

I used Cursor for building, ChatGPT for planning, and Google Stitch for the UI. The README was also drafted with AI help. This is an AI-assisted project developed under my direction.

Quality and verification

Verification so far has been manual: onboarding and AI chat checks, plus informal use by a few people. The public repository does not currently include automated CI test coverage.

Edge cases and failure handling

Login redirects and route gating exist in the code. Deeper failure-path documentation hasn't been written yet.

Known limitations

  • Simple demo model — not a full budgeting or bank-connected product.
  • Built with substantial AI assistance; I'm still studying the main application flows.
  • No major updates since the initial build.
  • Production user isolation has not been independently verified.

Next improvements

Learn the auth → check → calculation → result flow in more depth, rewrite the docs in my own words, and add the local tests to the repository.

Demo, repository, screenshots, and tests

Last verified date

July 2026 — repository and live demo reviewed.