ZyfinBot — Global personal finance SaaS

Problem

Household expense tracking dies on friction: any app that requires opening it, navigating to a form and filling five fields tends to have a high churn. The chat window people already live in is the fastest capture surface with low adoption cost — hence why this project focus on first-capture through Whatsapp.

Besides all of that, this project covers other pain points the founder currently has: global financial management (not only in one single currency and one single country) and financial consultants grant access for specialist help.

Approach

ZyfinBot is a Ruby on Rails application whose WhatsApp webhook lives in-process rather than as a separate service, gradually growing from a bot into a full multi-tenant financial app with a Hotwire web UI and a Flutter client.

Key architectural decisions:

• Official Meta WhatsApp Cloud API only, 1:1 with the bot — no unofficial clients, no ban risk.
• No LLM to start. A deterministic parser handles a one-line shorthand (02/06 Mercado Alimentação 130,28) and falls back to WhatsApp interactive buttons and lists, so behaviour is testable with golden tests instead of prompt tuning.
• Money as integer minor units with FX snapshot-on-write: each transaction stores the original currency plus the base-currency amount, rate, source and rate date, so historical reports never drift as rates move.
• Tenant-per-family isolation, starting on embedded SQLite for a zero-cost MVP and migrating to self-hosted PostgreSQL with Row-Level Security.
• Auth that assumes possible phone loss: email and password with mandatory 2FA, satisfied by either WebAuthn passkeys or TOTP — no SMS, no WhatsApp OTP.
• Infrastructure as code end-to-end — development environment on a Raspberry Pi k3s cluster behind a Cloudflare Tunnel, future production enviornment in the Cloud as a Terraform-provisioned deployed with Kamal 2, images built and pushed to GHCR by GitHub Actions.

Impact

• v0.1.0 foundation running locally: verified Cloud API webhook with HMAC signature checking, idempotent message handling keyed on the WhatsApp message id, expense parsing and persistence, all covered by RSpec golden tests.
• Delivery split into thin vertical slices — ingest, budgets, reports, Postgres/RLS migration, web UI, Open Finance — each independently shippable.
• Open Finance modelled provider-agnostically from the start, so Pluggy in Brazil and GoCardless/TrueLayer in Europe are configurable with no schema migration dependency.