Super Individual Contributor — I use AI as an operating system
to deliver the output of an entire team.
From diagnosing the problem → to the end-to-end solution → to the measurable result.
const vitor = {
role: "Software Engineer · Applied AI (Agents, LLMs & Automation)",
experience: "12+ years building and scaling SaaS platforms",
location: "Curitiba, Brazil 🇧🇷 · open to relocation to São Paulo",
currently: ["Pixel Educação", "SARCORPS (co-founder)", "ClearSeg (co-founder & CTO)"],
previousLife: "Air Traffic Controller @ DECEA / CINDACTA (2007–2022)",
motto: "Zero margin for error. Ship it anyway.",
buildsWithAI: [
"autonomous agents (Telegram / WhatsApp) with custom guardrails",
"in-product AI copilots grounded on each client's business data",
"LLM + Whisper summarization pipelines running every 3 minutes",
"AI fleets governed as code, multi-tenant and isolated",
],
};| 🎯 What I shipped | 📊 The number |
|---|---|
| Conversational AI agent operated live at a founders workshop (Sebrae) | 70+ concurrent users (~103 total) · 2,411 messages in 123 sessions — in minutes |
| In-product AI Copilot — natural-language Q&A over each client's business data | Sole author of the AI layer (OpenAI + Vercel AI SDK, context grounding) |
| AI summarization pipeline (automation + LLM + audio transcription) | ~700 automated analyses / day, running every 3 minutes |
| WhatsApp-group analytics SaaS (Next.js + Node/Express) | 3.6M+ interactions across ~2,900 groups |
| AI fleet as code (Tailscale + supervisor agent) | 19 governed agents · 9 servers monitored · isolated multi-tenant infra |
| Event-driven core platform (Node/TS + PostgreSQL) | ~400 production deploys with a 165-test suite green ✅ |
| Campaign & email pipeline | broadcast delivered to ~26.6k contacts |
| "Funcionário de IA" mentorship — created & taught from scratch | beta cohort rated 9.0 / 10 |
flowchart LR
A["📥 Real business problem"] --> B["🔎 Diagnose<br/>data + workflow"]
B --> C["🤖 Agent / Copilot / Pipeline<br/>LLM · RAG · MCP"]
C --> D["🛡️ Guardrails<br/>anti-injection · abuse gate · RLS"]
D --> E["🚢 Production<br/>tests · deploys · observability"]
E --> F["📈 Measurable result"]
F -.-> A
Guardrails are not decoration. The trust boundary is the server, not the UI — RLS,
SECURITY DEFINER, rate-limiting and PII sanitization ship with the feature, not after it.
- 🧩 Pixel Educação — Lead engineer of the event-driven core platform: multi-provider payment webhooks, lead attribution, automations + a fleet of autonomous AI agents for go-to-market, onboarding and student diagnosis.
- 🛟 SARCORPS — Co-founder & full-stack co-lead of a mission-critical Search & Rescue (SAR) operations platform for the aviation/defense sector (international IAMSAR standard). Nuxt 3/4 + Vue 3 + TypeScript + Supabase, with an in-app LLM assistant and a Python geospatial/simulation engine. Owner of data security, i18n (pt/en/es) and WCAG accessibility.
- 🧾 ClearSeg — Co-founder & CTO of an AI-powered recurring billing platform for insurance brokers: AI invoice reading (classification + extraction), white-label automated billing and portfolio management.
- 🎓 "Funcionário de IA" — Creator & instructor of an applied-AI mentorship: I translate agents, automation and LLMs into practical use for non-technical audiences.
I run a YouTube channel (in 🇧🇷 Portuguese) where I build AI agents, automations and dev-tooling live and unfiltered — including the parts that break. Real agents, real bugs, real fixes.
🤖 This list updates itself daily via GitHub Actions — the README is an automation too.
For 15 years (2007–2022) I was an Air Traffic Controller / Operator at DECEA / CINDACTA, the Brazilian airspace control authority — an environment of extreme pressure, zero margin for error and decision-making under uncertainty.
That's where the ownership discipline came from. Today I point it at production systems and AI:
✈️ separation minima → guardrails & rate limits
📡 radar vectoring → observability & tracing
🆘 emergency handling → incident response
🗣️ standard phraseology→ executive communication




