Clinic Management SaaS

The Problem

Brazilian medical clinics run on a patchwork: one tool for the calendar, another for records, a spreadsheet for billing, WhatsApp for everything else. Nothing talks to anything. And clinical data carries LGPD obligations that most of that patchwork quietly ignores.

My Role

Founder and Principal Engineer. Product, architecture, implementation, deploys. Every decision is mine, which means every decision has to be one I can still defend a year from now.

What I Built

Multi-tenant core

A Rails 8 monolith with PostgreSQL Row-Level Security. Tenant scoping lives in the database, not in application code, so a forgotten where clause can't leak one clinic's patients into another's. An in-database transactional event bus keeps side effects consistent with the writes that triggered them.

Server-driven SPA

Inertia.js with Vue 3 and TypeScript. No separate API layer to keep in sync, no duplicated validation, but still a fully typed frontend on a shadcn-based design system. The stack most people reach for here would have been two codebases; this is one.

The product surface

Scheduling, electronic medical records, care orchestration, patient self-booking, subscription billing. All of it in one place because that's the entire point.

Deploys

Kamal, single command, database provisioning through zero-downtime container swap. Shipping should not be an event.

Engineering Approach

Every non-obvious decision is written down as an ADR — 24 and counting — and the domain is backed by 170+ RSpec suites. Not for ceremony. A solo founder's codebase has no one to ask, so the codebase has to answer for itself. That turned out to matter twice over: it's what makes the project navigable for AI agents too.

Outcome

In production and running clinics. Still the codebase I reach for when I want to show what I actually think good software looks like.