Back
Case studyIn production2025 —

Zoan

Autonomous AI agent for Mercado Livre sellers

Zoan is a multi-tenant SaaS that behaves like an autonomous AI agent. I conceived, architected and built the product from scratch — from the first line of code to a production deploy.

zoan.com.br

The problem

Mercado Livre sellers lose hours on repetitive tasks: answering questions, creating and publishing listings (category, attributes, photos, copy), tracking orders and shipping, managing stock and pricing. Existing tools are fragmented and demand constant manual work.

The solution

A guided, autonomous pipeline. From a single photo and a short description, the AI writes an optimized title, professional copy and 5 studio photos — and publishes the listing on its own. Then it keeps working: answers buyers, monitors orders and centralizes everything in a mobile panel.

Listing publishing

AI writes title, copy and studio photos and publishes automatically.

Questions & answers

Answers buyers via AI, autonomously.

Order management

Monitors payment and shipping in real time.

Post-sale & shipping

Automatic messages at every step.

Inventory

Automatic sync, no spreadsheets.

Pricing & promos

Suggestions, with confirmation before touching money.

The product's golden rule

Anything reversible, the agent does on its own. Anything that moves money, it asks for confirmation. Autonomy with safety — the user decides on what truly matters.

Where AI comes in

  • Text with Gemini 2.0 Flash (primary) and GPT-4o-mini fallback.
  • Professional studio photos generated by AI (OpenAI gpt-image-1).
  • Confidence-based decisions: ≥ 80% auto-answers, ≥ 85% auto-publishes.
  • Mercado Livre events via signed webhooks, processed through queues (Redis + BullMQ).

Stack & architecture

NestJS 11Next.js 15TypeScriptPostgreSQL + TypeORMRedis + BullMQGemini 2.0 FlashOpenAI gpt-image-1Docker / OCIMercado Livre API v2

My role

From idea to a product in production: I defined the vision, designed the multi-tenant architecture, integrated the Mercado Livre API (OAuth + webhooks), built the backend (NestJS) and frontend (Next.js), modeled the database, implemented the credits + subscription model (Stripe) and shipped it with Docker. All with AI as an engineering lever — in record time, with clean and secure architecture.

Want a product like this?

I can take your idea from paper to production. The first conversation is no strings attached.