Otávio Costa

Client demos & steel analysis

IBM

IBM builds software and cloud solutions for enterprise clients. I worked there on two teams.

My work

On the Code Bakery team, I built client demos and a real-time app for Campus Party. Later, on the Gerdau project, I was tech lead on a system that reads technical steel specifications and calculates production risk.

Technology stack

Pick a technology to see what I used it for.

What I worked on

Pick a team to see its projects.

Client demos that reached production

IBM needed working demos to show enterprise clients what IBM Cloud could do. I built demos and proofs of concept and presented them with the team. Two of them went on to production: a chatbot that answers questions about FGTS for Caixa, and a security tool for Bradesco.

Two demos became systems the clients used in production.

How I kept demos fast to build

I built boilerplate projects in Angular, React and Ionic, and a UI library based on IBM's Carbon Design System. New demos started from them, so the team spent its time on the client's problem instead of setup.

The Bradesco tool was a crawler that used IBM's vulnerability detection to flag potential threats.

Connecting Campus Party visitors with IBM experts

At Campus Party, IBM wanted visitors to reach the right IBM specialist quickly. The team built a progressive web app for this, and I owned its WebSocket architecture and code.

Through the app, we made contact with 250 leads during the event.

How the live connection worked

I built the real-time layer with Socket.IO, so visitors and experts saw updates as they happened, without refreshing the app.

The app was built with Ionic, so it ran in the browser and on phones.

Reading technical steel specifications

Gerdau analyzed technical steel specifications by hand to judge whether a product could be made. As tech lead, I led the team that built an OCR and microservices pipeline with IBM's OCR and AI technologies to pull text, chemical elements and diagrams out of those documents.

The specifications became data the rule engine could calculate with.

What the pipeline extracted

Each document went through OCR, then separate NestJS services extracted the text, the chemical elements and the diagrams.

Calculating production risk

With the data extracted, Gerdau needed the production risk for each steel product. I owned the architecture and code of the graph and rule engine that calculated it from the chemical composition.

Analysis went from 13 days of manual work to 3 hours.

How the rule engine fit in

The rule engine applied Gerdau's manufacturability rules to the chemical elements extracted from each specification and returned the production risk.

Gerdau used the system in production, and the project became an IBM global success story.