Skip to content
Vertex Stack
Menu

Two build engagements per year

MicroMoneyMinded

Fintech SaaS

Behavioral budgeting, bank sync and AI insights in one PWA.

Problem
Budgeting apps record what people spend but not why, so small repeated purchases go unnoticed. Most tools are either passive spreadsheets or bank dashboards that show history without changing habits.
Outcome
A subscription PWA that pairs need-vs-want journaling with automatic transaction capture from Plaid, CSV, statement import and receipt OCR, plus budgets, goals, bills, debt planning and AI insights. It keeps working offline and syncs when the connection returns.
Role
Designed and built by Adil Mir
End-user pages
~20
Admin pages
~16
Database models
30+
REST endpoints
100+
Background job processors
~15
Tiered rate limiters
5

The problem

Most budgeting apps record what people spend but not why they spend it. Small, repeated, unconscious purchases (a $6 coffee, a $25 Friday takeaway) slip past and add up. Existing tools tend to be either spreadsheet-like and passive, or bank-aggregation dashboards that show history without changing habits.

MicroMoneyMinded is inspired by the Micro Money Minded financial philosophy. It pairs a daily finance journal with automated tracking and pattern detection, so the small leaks become visible.

What I built

I designed and built MicroMoneyMinded as a full-stack subscription SaaS, delivered as a mobile-first Progressive Web App. It combines a daily finance journal with automatic transaction capture (Plaid bank sync, CSV import, statement import and receipt OCR) and AI insights that point out spending leaks and suggest habits to close them. The app has about 20 end-user pages and an admin console of about 16 more.

Core modules

  • Dashboard. Budget progress, recent activity, active goals, journal streak, upcoming bills, AI insights and curated deals on one screen.
  • Transactions. Income, expense, transfer and savings types with need, want and savings buckets, tags, attachments and CSV import and export.
  • Budget. The 50/30/20 or 50/10/40 method, 30+ default categories, reusable templates, and month-to-month copy and comparison.
  • Goals. Target tracking with 20 icons and 15 colors, auto-contribution schedules, interest projections and a confetti animation on completion.
  • Bills. Recurring bill templates with 5 frequencies, a calendar view, mark paid, skip, pause and resume, reminders and overdue detection.
  • Debt. A payoff planner for 7 debt types, with snowball vs. avalanche simulation and full payment history.
  • Journal. Daily entries with a need/want breakdown, mood tracking, guided prompts in 7 categories and streak milestones.
  • Reports. 10+ report types, including spending by category, income vs. expenses, trends, budget vs. actual, savings rate and net-worth history.
  • Calculators. Compound interest, loan amortization and retirement projections, saved for reuse.
  • Deals. Curated Amazon-affiliate deals with filters, saved favorites, price-drop alerts and click attribution.
  • AI insights. Pattern detection, trend analysis, warnings and projections from a hybrid rule-based and LLM pipeline.
  • Micro-splurge detection. Configurable detection of small repeated purchases, with the projected annual and multi-year savings if the habit stopped.

Platform features

  • Installable on desktop and mobile, with standalone display and app shortcuts.
  • Offline support: IndexedDB caching, a background sync queue and automatic reconciliation.
  • Push notifications for bill reminders, goal milestones, journal prompts, deal price drops and budget alerts.
  • Bank statement import from PDF or image, with parsing for multiple bank formats.
  • CSV import with a column-mapping UI, preview, validation and undo.
  • Email-based two-factor authentication with 6-digit codes and brute-force protection.
  • Full data export and account deletion with cascade cleanup, built with GDPR in mind.
  • Per-user currency, locale and timezone, and light, dark and system themes.
  • An admin console for users, subscriptions, deals, support, notifications, analytics and system health.

Habit mechanics

  • Journal streaks with milestones at 7, 14, 30, 60, 90 and 365 days.
  • Goal notifications at 10, 25, 50, 75 and 100 percent, with a celebration on completion.
  • Budget alerts at 80 percent (warning) and 100 percent (exceeded).
  • A weekly summary email.

Subscription tiers

There are three tiers. Free keeps 90 days of transaction history and caps active goals and recurring bills. Plus removes those limits and adds custom categories, auto-contribution, the debt planner, statement import and a monthly receipt-OCR allowance. Pro adds Plaid sync for up to 5 accounts, AI insights and micro-splurge detection, month-to-month budget comparison and net-worth tracking.

Integrations

Each external service sits behind its own adapter with retries, rate-limit handling and webhook reconciliation.

  • Plaid for bank linking, cursor-based transaction sync, webhook-driven re-authentication and multiple accounts.
  • Azure Document Intelligence for receipt OCR: merchant, date, total and line items, each with a confidence score, plus a correction step for the user.
  • OpenAI (GPT-4o-mini) for natural-language insights, with 24-hour caching and transaction batching to control cost.
  • Stripe for Checkout, the customer portal, and upgrade, downgrade, cancel and reactivate flows, with subscription state synced from webhooks.
  • Firebase Cloud Messaging and Web Push for notifications on web, Android and iOS.

Architecture

Offline-first sync

A custom sync engine reconciles changes made while the device is offline. Each operation is written to IndexedDB, tagged for background sync and flushed in ordered batches when connectivity returns. The React Query cache is persisted locally, so data is on screen as soon as the app opens, before any network call.

Tiered feature gating

One middleware maps subscription tiers to feature sets. Gated routes declare what they need with requireFeature or requireAnyFeature, and the client reads the same mapping to decide where to show upgrade prompts. The server's enforcement and the client's UI come from one source, so they stay in step.

Hybrid AI insights

The insight pipeline has two layers. Deterministic, rule-based detectors find the patterns: repeated purchases, shifts in category share and budget pace. An optional LLM pass with GPT-4o-mini then writes the natural-language explanation. Results are cached per user for 24 hours, which keeps LLM spend predictable.

Background workers

Bull queues on Redis handle everything that does not need to happen during a request: goal milestones, streak achievements, budget alerts, weekly summaries, price-drop monitoring and savings reminders. About 15 job processors run these. Jobs respect each user's timezone and retry with exponential backoff.

Stripe as the source of truth

Billing state comes from Stripe, not from the app. Webhook handlers for successful and failed invoice payments, failed charges, and subscription updates and deletions keep the user record, feature access and email notifications in line with what Stripe reports.

Receipt pipeline

An uploaded receipt is first normalized, resized and sharpened with Sharp. Azure Document Intelligence then extracts structured fields with per-field confidence. Low-confidence fields are shown to the user for review before the app creates a linked transaction.

Security

  • Passwords hashed with bcrypt at 12 rounds.
  • Short-lived access tokens and rotating refresh tokens, with a client-side mutex so only one refresh runs at a time.
  • Email-based two-factor authentication.
  • Brute-force protection with progressive lockout.
  • Per-endpoint rate limiters for auth, password reset, 2FA and email verification.
  • Strict CORS, Helmet headers, NoSQL-injection sanitization and HTTP parameter pollution protection.
  • AES-256 encryption for sensitive fields at rest.
  • Cascade deletion of all user data on account closure.

What it demonstrates

  • Full-stack SaaS delivery, from authentication to billing to admin operations.
  • Subscription architecture: tiered feature gating, the Stripe lifecycle and webhook reconciliation.
  • Several third-party services (Plaid, Stripe, Azure OCR, OpenAI, Firebase and web push) working together behind consistent adapters.
  • PWA and offline engineering: service workers, IndexedDB, sync queues and background sync.
  • Event-driven background work on Redis-backed queues, scheduled per user timezone.
  • AI used where it helps: rules find the patterns, the LLM explains them, and caching bounds the cost.

Planning something similar?

Tell me what you're building. I reply personally within one business day.

Engagements from $50,000 · Two builds per year