Zaldo
An Indonesian personal finance app that sets bills aside first, then shows how much is safe to spend until the next paycheck.
- Role
- Creator & Engineer
- Status
- Live
- Stack
- Next.js
- TypeScript
- Fastify
- PostgreSQL
- Prisma
- TanStack Query
- Zod
- Turborepo
- TailwindCSS

The problem
A large balance says little when part of it is already promised to bills, debts, and savings.
What changed
Safe-to-spend money, upcoming payments, and purchase checks come from exact calculations, and an assistant explains them in chat.
Context
People in Indonesia often track money across several bank accounts, e-wallets, and cash, while bills, installments, and savings goals compete for the same balance. I built Zaldo as my own product to answer one question first: how much can I spend until payday without missing a payment?
Engineering decisions
Built a pnpm and Turborepo monorepo: a Next.js app, a Fastify API on PostgreSQL through Prisma, shared Zod contracts used by both sides, and a framework-independent finance engine. The engine works in exact minor units and owns safe-to-spend, forecasts, and spending insights, so the interface and the assistant never calculate money themselves.
My contribution
- Designed safe-to-spend as a total allowance until the next income, after mandatory payments and protected savings are set aside.
- Built a ledger with accounts, pockets, transfers, budgets, recurring bills, debts, commitments, and savings goals, with balances derived from entries instead of stored counters.
- Built a finance assistant that reads data through 27 allowlisted tools and the same API routes as the app; every change it suggests is a preview that the user must confirm.
- Kept arithmetic out of the language model: a dedicated calculator tool evaluates named steps exactly with BigInt fractions.
- Defined 117 versioned API endpoints with shared request and response schemas, and recovery for writes whose outcome is uncertain.
- Shipped it as an installable PWA with update prompts, optional push reminders, and a payment badge.
- Added receipt reading through a vision model that produces drafts for review, never saved transactions.
Trade-offs
- Financial data stays online-only, with no offline queue, so a write is never replayed twice without the user knowing.
- Zaldo does not connect to bank accounts: it cannot move money, at the cost of recording transactions by hand or through chat.
Lighthouse
- Performance
- 89
- Accessibility
- 100
- Best practices
- 96
- SEO
- 100
Mobile test of the home page, median of three runs, on . Scores vary between runs and devices.
Next project
Questions about this work?
Ask the assistant, or email me directly.