At Zoppy, a marketing automation platform for e-commerce brands, I went from mid-level engineer to leading the largest engineering chapter, about ten engineers, in two years.
Along the way I led the AI Engineering squad, created our WhatsApp Commerce AI sales agent, re-architected customer segmentation, overhauled our event-driven email pipeline and became the company’s reference for security and AWS cost.
Before that I co-founded Porta, an Open Finance credit-portability marketplace, where I owned product, design and engineering. I work AI-native every day with Claude Code and Codex, and I’m based in Belo Horizonte, Brazil.
Selected work
Open any project for the problem, what I did and the result.
WhatsApp Commerce An AI sales agent that sells to shoppers inside WhatsApp. I created it and own it.
- Problem
- Zoppy’s customers are e-commerce brands that already reach their shoppers on WhatsApp. The opportunity was to let shoppers buy in the conversation itself.
- What I did
- Created the product and lead it: a Python AI agent that uses LLM tool use to sell to shoppers, running on the Meta WhatsApp Business Platform.
- Result
- WhatsApp Commerce is a Zoppy product, selling to shoppers inside WhatsApp, and I own it end to end.
Event-driven email pipeline Overhauled the pipeline behind every email event: cheaper, faster and far more reliable, while traffic grew.
- Problem
- The platform sends millions of messages a month, and every email produces events that flow through an event-driven pipeline on SES, SNS, SQS, Lambda and Cassandra. Its cost, latency and error rate all had to come down, with traffic still rising.
- What I did
- Led the overhaul, moving the pipeline to direct asynchronous enqueueing, and led the retirement of a paid AWS deliverability service.
- Result
-
- −94%cost
- −92%latency
- −98%errors
Segment v2 Single-handedly re-architected customer segmentation, one of the platform’s core features.
- Problem
- Brands on Zoppy target campaigns and automations at customer segments, so segmentation sits at the core of the platform. It needed a new foundation.
- What I did
- Re-architected it on my own: new segmentation logic and architecture, built as a new library that I started and remain the top author of.
- Result
- Segment v2 shipped as the platform’s customer segmentation, designed and delivered by one engineer.
AI Engineering squad Led the squad that shipped Zoppy AI, then the Copilot, an MCP server and a customer-success agent.
- Problem
- Brands plan campaigns, build segments and write copy inside Zoppy. Each of those jobs was a place where AI could do the first draft.
- What I did
- Led the AI Engineering squad as tech lead. We built Python agents that draft multi-channel campaigns, build segments from plain-language requests and write email and SMS copy. As lead of the Features chapter I went on to lead the in-app AI Copilot, an MCP server for AI assistants and an LLM-driven customer-success agent.
- Result
- Zoppy AI shipped, and the squad’s work led into the Copilot, the MCP server and the customer-success agent that I went on to lead.
Porta An Open Finance marketplace for loan portability, licensed as a banking correspondent.
- Problem
- Open Finance made it possible to see a borrower’s loans and let other banks bid to take them over. Porta set out to turn that into a marketplace for consumers and companies.
- What I did
- Co-founded Porta and owned product, design and engineering. Consumers and companies connected their loans through Open Finance or entered them by hand, and received portability offers from banks. Built on NestJS, Next.js and React.
- Result
- A live credit marketplace with real customers, licensed as a banking correspondent.
Edward Give it a store’s URL and a one-line prompt, and it renders an on-brand video ad.
- Problem
- Could an agent turn nothing more than an e-commerce URL and a prompt into a finished MP4 ad that looks like the brand made it?
- What I did
- Built a multi-tenant API. Playwright and Gemini’s multimodal models extract the brand’s identity from its site; a LangGraph agent writes the ad as a JSON composition language; a Node Remotion worker renders it. One JSON Schema generates both the Pydantic and Zod types, and CI fails if they drift. Infrastructure is four AWS CDK stacks, using VPC endpoints instead of a NAT gateway.
- Result
- A working prototype that goes from URL and prompt to a rendered MP4, covered by around fifty test files.
Developer CLIs in Go Two command-line tools I wrote alone, in my language of choice since university.
- Problem
- Recurring engineering work, at Zoppy and on my own machine, is best done through tools with a clear interface.
- What I did
- Wrote two developer CLIs in Go as sole author: one for Zoppy’s engineering work and one personal toolkit.
- Result
- About 7,000 and 6,000 lines of Go respectively, both started in mid-2026.
Experience
-
Zoppy
Marketing automation for e-commerce brands. Belo Horizonte, hybrid.
-
– Present
Tech Lead, Features chapter
- Lead the company’s largest engineering chapter, about ten engineers: feature delivery end to end, code review, production incidents and release quality.
- Created and own WhatsApp Commerce, an AI sales agent selling to shoppers inside WhatsApp.
- Led the in-app AI Copilot, an MCP server for AI assistants and an LLM-driven customer-success agent.
- Own billing and payments: subscriptions, invoices, checkout, Pix and card payments across two payment providers.
- Overhauled the event-driven email pipeline: −94% cost, −92% latency, −98% errors while traffic grew.
- Ran a FinOps audit identifying savings of up to 18% of the AWS bill, and built and ran the company’s security audits, taking critical findings from 4 to 0.
-
–
Tech Lead, AI Engineering
- Led the AI Engineering squad that shipped Zoppy AI: agents that draft multi-channel campaigns, build segments from plain-language requests and write email and SMS copy.
-
–
Senior Software Engineer
- Single-handedly re-architected customer segmentation (Segment v2).
- Built queue-based messaging and data-sync workers on RabbitMQ and BullMQ, and integrations with about 20 e-commerce platforms and 12 ERPs.
-
–
Software Engineer
- Full stack features on NestJS and Angular for a platform sending about 22 million messages a month across email, WhatsApp and SMS.
-
-
Porta
Open Finance credit marketplace. Belo Horizonte.
-
–
Co-founder
- Owned product, design and engineering for a loan-portability marketplace licensed as a banking correspondent.
-
-
ioasys
Software studio. Belo Horizonte.
-
–
Software Developer Intern
- Joined through ioasys Camp, a competitive selection where my team won the final challenge.
- Built a digital-banking product for a fintech client across front end and back end with NestJS, Next.js and React.
-
Education
–
B.Sc. Information Systems
Universidade Federal de Minas Gerais (UFMG)
Average 87.6 out of 100. Thesis on transformer models for predictive maintenance at an industrial food producer, through a university–industry partnership. Coursework in Deep Learning for NLP, Machine Learning, Data Mining, Algorithms, Software Engineering and Cybersecurity.
Skills
Languages
- TypeScript
- Python
- Go
- SQL
Back end
- NestJS
- Node.js
- FastAPI
- REST APIs
- Microservices
- Payments
Front end
- React
- Next.js
- Angular
AI
- LLM agents
- LangGraph
- RAG
- LLM tool use
- Prompt engineering
Data and messaging
- PostgreSQL
- MySQL
- Apache Cassandra
- Redis
- OpenSearch
- RabbitMQ
- Event-driven architecture
Cloud and operations
- AWS
- AWS CDK
- Terraform
- Docker
- CI/CD
- Grafana
- Sentry
Leadership
- Technical leadership
- Software architecture
- Product management
- Product design
Spoken
- English, full professional
- Portuguese, native