Decision-Support Application · 2026 · Live

RetireCheck

People need a credible “will my money last?” answer without advisor-grade onboarding friction.

Problem worth solving

People approaching retirement need a credible answer to one question: will this plan last? Advisor-grade tools are heavy; spreadsheets hide assumptions. RetireCheck gives a structured decision-support path without accounts, paywalls, or advice claims.

Intended user

US pre-retirees and planners exploring Social Security claiming ages, portfolio longevity, taxes, RMDs, and Medicare IRMAA context — people who want clarity before they talk to an advisor, not a substitute for one.

RetireCheck readiness gauge on the results page — Results surface: readiness score from 1,000 Monte Carlo scenarios — also reachable via a one-click sample plan.

Constraints that shaped the solution

Domain math must be correct and testable — not UI-owned
No sign-up friction; estimates must carry clear disclaimers
Deployable monorepo: Next.js frontend + .NET calculation API
Production CI — not a demo that stops at happy-path UI

Architecture decision

Financial math lives in a pure C# domain layer. An ASP.NET Core API exposes that domain. Next.js owns the wizard UX and charts, and proxies to the API so the browser never talks to calculation code directly. Tests sit against the domain and API — not only the UI happy path.

Governing rule: all math lives in the C# domain layer — never in React components or controllers.

RetireCheck system architecture diagram — Next.js → API proxy → ASP.NET Core → RetirementCalculator.Domain.

Tradeoffs

Accepted: two runtimes (Node + .NET) for a clean domain boundary
Rejected: putting Monte Carlo or tax logic in the browser
Delayed: accounts, saved plans, and advisor workflows until evidence justifies them
Excluded: financial advice, recommendations that imply fiduciary guidance

Monte Carlo fan chart on RetireCheck results — Annotated product surface: P10–P90 fan chart + median path after architecture was fixed.

How AI was used — and what stayed human

Cursor accelerated UI scaffolding, charts, and boilerplate under written project rules. Human judgment owned domain boundaries, test strategy, disclaimers, deployment topology, and what never to ship. Speed was not the achievement — correctness and production readiness were.

Intent-driven Cursor workflow diagram — Context → Loop → Ship: engineer decisions, AI assistant keystrokes, pipeline verification.

Privacy, security, correctness, accessibility

No account required; inputs are session-scoped for calculation
Calculation API isolated behind Next.js proxy
xUnit domain/API tests + GitHub Actions CI
Explicit “not advice” disclaimers on results
Keyboard-reachable wizard steps and readable results chrome

What was delivered

Four-step planning wizard with progressive inputs
One-click sample plan for first-time visitors
Animated score gauge and shareable results summary
Monte Carlo fan chart (P10–P90 band + median)
SSA FRA lookup and tax / RMD / IRMAA context
Live app on Vercel + API on Render

Lessons

Write domain rules before accelerating UI with AI.
Put correctness-critical logic where tests are natural.
Disclaimers and empty/error states are product features.
Document the loop: Build → Validate → Improve → Document → Share.

Limitations and future considerations

Estimates are educational, not advice. Broader tax jurisdictions, saved plans, and advisor export would only be added if real usage evidence justified the complexity. App-store packaging and monetization are explicitly out of scope for this case study.

Full delivery narrative: AI in Action: Building RetireCheck with Cursor.