By the numbers
- paying subscribers
- 500+
- MRR at peak
- $50K+
- less manual support
- ~60%
- per feature, down from 1–2 weeks
- 2–3 days
Overview
A SaaS built on top of the YouTube APIs: AI-powered comment analysis that filters toxic comments, detects spam ads, finds leads, and lets channel owners chat with AI about their audience. Payments run through Hotmart, a Brazilian platform for digital products.
I built and ran it as the primary engineer, owning it end to end: product, backend, payments, deploys and support. It grew from zero to 500+ paying subscribers and $50K+ MRR at peak, and it is live in low-maintenance mode.
What I built
- 01Grew it from zero to 500+ paying subscribers and $50K+ MRR at peak as the primary engineer.
- 02Backend on GCP (Cloud Run, Cloud Functions, Pub/Sub) alongside AWS Lambda and Vercel, with queues and scheduled jobs designed to stay idempotent under retries and concurrent writes, so growth did not create duplicate charges or corrupted data.
- 03RAG support assistant (OpenAI, LangChain, Qdrant) that cut manual support handling by ~60%.
- 04YouTube APIs, Hotmart payments and email/analytics wired up behind audited REST APIs.
- 05Spec-first with AI coding agents: I wrote acceptance criteria, let the agent draft, reviewed its plan and corrected wrong architecture calls before merging. Features shipped in 2–3 days instead of 1–2 weeks.
Case studies
Idempotent from day one
Problem
A paid SaaS with queues and scheduled jobs across GCP and AWS, where retries and concurrent writes could double-charge users or corrupt data as it grew.
What I did
- Designed every queue consumer and scheduled job to be idempotent and safe under retries and concurrent writes.
- Owned product, backend, payments (Hotmart), deploys and support.
- Added a RAG support assistant (OpenAI, LangChain, Qdrant).
Result
500+ paying subscribers, $50K+ MRR at peak and ~60% less manual support.
Stack
- React
- TypeScript
- Tailwind CSS
- Node.js
- Express
- GCP Cloud Run
- Cloud Functions
- Pub/Sub
- AWS Lambda
- Vercel
- OpenAI
- LangChain
- Qdrant
- Hotmart