Português
Gabriel Vieira

Gabriel Vieira

CTO, US @ Ubivis

Machine learning for the factory floor: physics-informed models, rare-event detection and digital twins.

At a glance

  • 5+years teaching science and AI
  • 6people on the AI team I lead
  • 1,000+educators reached by EducaIA
  • 5languages I speak

About

I build machine-learning systems that run in production on factory floors. At Ubivis, an industrial AI company I co-own, I lead the AI modelling team: five ML engineers and a product manager working on defect detection, process models for steelmaking and digital twins connected to PLC, SCADA and MES.

I trained as a physicist, and it shows in how I work: I prefer models that respect the physics of the process, because plant data alone is rarely enough. My M.Sc. at UTFPR is on physics-informed machine learning for industry.

Teaching is the other half of my career. I taught Computer Vision at PUCPR and designed EducaIA, an AI bootcamp for more than 1,000 Brazilian educators.

Now

Leading the AI modelling team at Ubivis, working on physics-informed models for heavy industry and on the company's growth in the United States, while finishing my M.Sc. at UTFPR.

Principles

  1. Physics first, data second.

    Plant data rarely covers the whole operating range. Models that respect the process keep working where the data runs out.

  2. No model ships without review.

    Quality gates, safety checklists and an ethical-use policy are part of the product, not paperwork.

  3. If the operator cannot trust it, it does not exist.

    A model is only useful when the people on the floor understand what it says. Explaining it clearly is part of the engineering.

Experience

  1. 2024 — Present

    CTO, US · Ubivis

    Co-owner since 2024, responsible for technology in the United States. Lead the AI modelling team and own its budget, vendor choices and ROI models. Built the company's AI governance framework: quality gates, safety checklists and an ethical-use policy, so no model reaches production without review.

    • Team leadership
    • AI governance
    • MLOps
  2. Feb — Oct 2024

    Data Scientist, AI · Ubivis

    End-to-end ML pipelines for industrial clients, from data acquisition to containerized deployment.

    • Python
    • PyTorch
    • Docker
  3. 2023 — 2024

    AI Resident · SENAI Hub of AI

  4. Oct — Dec 2025

    Lecturer, Computer Vision (undergraduate) · PUCPR

  5. Instructional design

    EducaIA bootcamp · Santander Open Academy / DIO

    Scripts, prompt templates, rubrics and quality checklists covering ethics and bias detection, for an AI course taken by 1,000+ educators.

Selected work

  • Rare-event defect detection in robotic welding

    A binary classifier for defects that almost never happen, trained on extremely imbalanced sensor data with 700+ time-series features per weld cycle. It runs end to end, from acquisition to a containerized deployment. The project, delivered by Ubivis, was a Top 3 finalist of the FINEP Innovation Award 2025 (Digital Transformation of Industry).

    • Time series
    • tsfresh
    • Imbalanced learning

  • Physics-informed models for steelmaking

    Neural networks that combine process physics with plant data for basic oxygen and electric arc furnaces, where data alone does not cover the operating range.

    • PINNs
    • PyTorch
    • Process modelling

  • ML pipelines over 170 GB of sensor data

    Polars with Zstandard-compressed Parquet, disk caching, adaptive undersampling with MiniBatchKMeans and parallel feature extraction, sized to run on a single machine.

    • Polars
    • Parquet
    • Data engineering

  • Internal LLM platform

    Retrieval-augmented generation with reranking and a two-pass draft-and-verify pipeline, built on LangGraph, with Telegram, Alexa and web interfaces.

    • LangGraph
    • RAG
    • LLMs

Awards & talks

  1. Sep 2026

    Peer-reviewed paper and talk · ROG.e 2026, Rio de Janeiro

    Physics-informed machine learning for risk-based inspection of oil and gas equipment, presented on stage on September 24.

  2. 2025

    Stage presenter for Ubivis · GITEX Europe, Berlin

    Presented Ubivis's industrial AI platform to a global audience.

  3. 2018

    Valedictorian · Instituto Cultural Brasil Estados Unidos (ICBEU)

  4. 2017

    Brazilian Youth Ambassador · U.S. Department of State

    Selected for the program's leadership training and exchange in the United States.

Education

  1. Sep 2026

    IELTS Academic — overall 8.0 · CEFR C1

    Listening 8.5 · Reading 8.5 · Speaking 8.0 · Writing 7.0

  2. In progress

    M.Sc. in Computer Science (professional) · UTFPR

    Physics-informed machine learning for industry.

  3. 2025

    Specialization in Information Technology in Education · UEL

  4. 2023 — 2024

    B.Sc. and Teaching Degree in Physics · Universidade Estadual de Londrina

    Research in computational cosmology and X-ray microtomography.

Languages

Portuguese (native) · English (IELTS Academic 8.0) · German (B2) · Spanish · French