Effable
A personal finance app I built solo, from data model and API design through to LLM cost and abuse controls.
TL;DR
- Outcome
- Shipped solo, from data model and API design through to LLM cost and abuse controls
- Role
- Founder and solo developer
- Context
- AI-backed personal finance app, available on iOS since late 2026
Context
Effable is a personal finance app for iOS and Android. People log spending in seconds, by typing a phrase, by voice or with a few taps, and get plain-language answers about where their money goes, including from an AI assistant that answers questions about their own records. It doesn’t link to bank accounts: financial privacy comes first. I built it alone, so the design had to stay simple enough for one person to keep correct, and the AI features had to stay cheap to run and hard to abuse.
Architecture
- Mobile app. React Native and Expo, one codebase for iOS and Android.
- API. One NestJS REST API in TypeScript; chat replies stream to the app as they are written.
- Database. PostgreSQL through TypeORM, with hand-written migrations.
- AI. Anthropic’s API behind a provider seam, reached only through a usage guard that every AI call passes.
- Outside services. Subscriptions, push notifications, email and exchange rates come from outside services the API calls.
Decisions
Balances derived on read. I don’t store balances; every balance is computed from the records when it is read. Backdated edits, imports and deletes just work, with no reconciliation jobs to keep a stored total in sync.
One authorisation seam for shared accounts. Accounts can be shared between people, and the alternative was an access check in every query. One missed check would leak someone’s finances, so all multi-tenant access goes through a single check.
Patterns first, the model second. Typed entries are tried against simple patterns, and the model is called only when they fail; calling it for every entry would cost more on every log. The assistant can only read, only the signed-in user’s own data, and every call passes layered limits and a per-user spend budget.
Result
Effable has been available on iOS since late 2026. I built it solo, from data model and API design through to LLM cost and abuse controls.
My role
I founded Effable and was its only developer. I built the core first, then adopted Claude Code and built my own setup for it: an agentic workflow and a team of agents with layered instructions and subsystem skills, delivering whole feature slices per commit. It sped up finishing the app a lot.