pten
Available for projects & opportunities

I'm an
AIEngineer&ProductBuilder
— from architecture to a production deploy.

I build and operate Konektor, the platform consultancies use to run entire portfolios of Mercado Livre sellers — today 56 accounts and R$ 30.2M in GMV over 12 months. From the first sketch to the daily operation.

What I do

From idea to a live product.

I take an idea, sometimes just a sentence, and turn it into a real, working product: from vision to architecture, from the first sketch to a production deploy.

I don't ship just code: I ship the product, with product vision and architecture built to grow. I designed and built Konektor from scratch, and today it runs in production for 12 consultancies.

seller accounts under management
56
listings managed by Konektor
42k
GMV over 12 months
R$30.2M

Got an idea in mind? The first conversation is no strings attached.

Chat on WhatsApp

  • AI products

    Apps and agents powered by generative AI: automated financial auditing, intelligent classification, decision automation and autonomous flows wired to frontier models (Gemini, GPT, Claude).

  • Automation & integrations

    Connecting APIs, marketplaces and systems to remove manual work: operational management, data pipelines and ERP-to-platform integrations that run on their own.

  • Software engineering

    Performant frontend, scalable backend and well-modeled data. B2B platforms, operations dashboards and internal systems, with clean, secure architecture built to evolve.

How a project moves

  1. 01

    Idea & vision

    Getting to the real problem and designing the answer: what matters, what scales, what creates value.

  2. 02

    Architecture & build

    Building by directing AI agents: clean, secure architecture, fast, without trading away quality.

  3. 03

    Shipped & live

    Deploy, monitoring and continuous evolution. Something that works in the user's hands, not a prototype.

Projects

Products I designed and built, from concept to production code.

konektor.com.br/demonstracao
Konektor's public demo, with fictional data: one master product linked to eight listings

Konektor's public demo, with fictional data.

Live2026

Konektor

The platform consultancies use to run whole portfolios of Mercado Livre sellers

The problem
A consultancy looks after many seller accounts at once, and each one needs stock, listings, promotions and finances kept current.
What I did
I defined the product, directed the agents that built it and I answer for it in production.
Where it is today
Live: 12 consultancies, 56 seller accounts and 42k managed listings. R$2.78M in GMV in August 2026 and R$30.2M over the last 12 months. An AI agent operates accounts under human approval.
Visit Konektor
Claude · KonektorIllustrative example

Which accounts in the portfolio are about to run out of stock this week?

Three accounts have products with under 7 days of stock. On the Aurora account, 2 listings sell faster than the balance can hold.

I propose pausing those 2 listings until restock. Nothing changes without your approval.

A sample conversation, not a capture.

Live2026

Konektor MCP

Talk to Konektor in plain language, through Claude

The consultant asks a question or requests an action in plain language; the agent investigates the seller portfolio and proposes the operation. Every write waits on human approval.

brooking.rosiak.com.br
Brooking's public page, showing the day's agenda with sample data

Live2026

Brooking

Personal agenda operated by AI agents

The agents book, postpone and schedule on their own, over MCP; every write stays on the screen, which is where I check what happened.

Open the site
zoan.com.br
Zoan's public page

Live2026

Zoan

Konektor's successor

In-house management software designed to predict market patterns and run the whole company. The foundation is built; the predictive capability is the goal, and it is not finished yet.

NestJS · Next.js · TypeScript · PostgreSQL · Redis · BullMQ

Read the case studyOpen the site
rosiak.com.br
This site's first screen, with the shader smoke and the 3D object

Live2026

Portfólio

This site, built by directing agents

A WebGL first screen: shader smoke and a real-time 3D object that answer the cursor and the scroll. Bilingual, PT and EN.

Next.js · React Three Fiber · GLSL

Code
Claude · InstagramIllustrative example

Post the new collection carousel tomorrow at 10am, with the caption we approved.

Carousel with 5 images assembled and caption checked. Scheduled for tomorrow, 10:00.

A sample conversation, not a capture.

Live2026

Instagram MCP

Publishes to the brand's Instagram by talking to Claude

Image, carousel, story or reel, asked for in a conversation with Claude and published to the brand's Instagram.

zcel.com.br
Zcel store home page

Live2026

Zcel

B2B Apple-parts store for a partner

Apple repair parts for an industry partner, with tiered pricing and access restricted to registered businesses.

Open the site
In development
Nothing public to show yet.

In development2026

Log Pose

My personal agent

A personal AI agent that will one day run Brooking in place of Claude Code. Still in progress.

rosiak-classifier.onrender.com
Imu Email Classifier's public page

Live2024

Imu Email Classifier

Real-time email classification with AI

Bulk email classifier powered by generative AI that separates what is urgent from the noise and speeds up response.

React · Python · Flask · Gemini

Open the site

What I take on

Whichever of the three, I start with scope: what is in, what is out, and the deadline.

  1. A new product

    I take the idea, set the scope and direct the AI agents that build it, checking every delivery until the product is live.

    Examples: Konektor, Brooking, this site

  2. What is already live

    I run what is in production: fixes, new releases and the product's evolution after launch.

    Examples: Konektor, which I run today

  3. Projects for your business

    Stores, integrations and custom tools for a business, from scope to delivery.

    Examples: Zcel, Instagram MCP

About

AI as a lever, product as an obsession.

Into tech since age 6. What started as curiosity became an obsession with product and engineering.

AI as a team, not as decoration

I run several workstreams at once by orchestrating AI agents: I split the work, set what each stream owns, review what comes back and answer for the whole. That is what lets me decide scope, cut what will not fit the deadline, and carry the product all the way into operation.

Since the age of 6, I've been taking things apart, testing and building on the computer. That curiosity never stopped: it became study, it became practice, and today it's how I create.

I'm studying Computer Science at UniRitter/FAPA, in my 6th semester, expected to graduate in 2027. I work end-to-end as a software and product engineer, today building and operating Konektor in production, focused on architecture and on shipping something that truly works.

The desire to know is the driving force behind all science.

Profile

Product
Konektor, platform in production
Scale
56 accounts · 42k listings
Method
AI agent orchestration
Studies
Computer Science, UniRitter/FAPA

Tools

  • TypeScript
  • React
  • Next.js
  • Node.js
  • Python
  • Tailwind CSS
  • PostgreSQL
  • Docker
  • OpenAI
  • Claude
  • Three.js
  • Git

What's next

Zoan is the next step: Konektor's successor, with market-pattern prediction and whole-company management, still in development. Open to product engineering opportunities and partnerships around directing AI agents.

Let's talk

A project, a partnership or an opportunity. Pick the channel you prefer; I reply fast.

Or directly