ML Engineer

Dimitre Oliveira Portfolio

AI Engineer at Intuition Machines Inc.

Kaggle Grandmaster and Google Developer Expert (ML) working as an ML engineer at Intuition Machines; notebooks and GitHub span TensorFlow/Keras production pipelines through to modern generative AI (Gemini, Vertex AI, Stable Diffusion fine-tuning) apps.

TensorFlowKerasGCPGenerative AI

Location

Campina Grande, State of Paraíba, Brazil

Companies

Intuition Machines Inc.Virtus, UFCG

Skills

TensorFlowKerasGCPGenerative AI

Notable

  • Kaggle Notebooks Grandmaster (ranked 2,635 of 60,583)
  • Google Developer Expert (GDE) in AI and Cloud

What makes it work

A breakdown of the choices that make this portfolio stand out.

1

Google Developer Expert status as third-party validation

GDE status is awarded by Google based on community contribution and depth, independent of any employer — a second institution vouching for the same skill set as his Kaggle rank, alongside a Kaggle Grandmaster tier that is itself earned through a public, competitive, and fully auditable track record rather than self-reported experience.

2

Demonstrated range from classic deep learning to current GenAI

Shipping projects across TensorFlow/Keras and newer Gemini/Vertex AI/Stable Diffusion work shows the portfolio is actively maintained, not stuck on 2019-era tutorials.

3

Full small-app builds, not just model notebooks

Building and shipping small end-to-end apps (a Streamlit PaLM app, a deployed generative Hangman game) proves he can wrap and ship a model, closer to the real ML engineer job than a notebook that ends at a metric.

4

Dual-platform credibility (Kaggle competitions plus GCP tooling)

Competing at Grandmaster level is a pure-signal, adversarial proof of modeling skill, while the Vertex AI/GCP-flavored generative work shows the separate, complementary skill of shipping on a specific cloud platform's production tooling — most candidates show only one side of that pair.

What ML Engineers can take from this

Specific, actionable tips to apply to your own portfolio — no generic advice.

  • Pursue independent technical recognition (GDE, MVP, etc.) alongside competition results — two different institutions vouching for you is stronger than either alone.

  • Keep at least one recent project on this year's frontier stack so the portfolio does not read as frozen.

  • Ship the model behind a real interface at least once — deployment beats a bare notebook.

  • Write a short post on the deployment decisions for every project, not just the modeling choices.

  • If you compete on Kaggle, pair that adversarial proof of modeling skill with at least one project that demonstrates production/cloud-platform fluency — the two together cover what most single-track portfolios miss.

Projects

Grandmaster-tier Kaggle notebooks show the modeling; a generative Hangman game shows the shipping. Dimitre Oliveira keeps both in view, pairing competition-grade notebook work with small deployed GenAI apps — closing the leaderboard-to-product gap most ML portfolios leave open.

Toolintermediate2-4 weeks

Deployed Generative AI Apps: Streamlit PaLM App & Generative Hangman

TensorFlowKerasPaLMStreamlitGCPVertex AI

Two small end-to-end generative AI apps built and shipped beyond notebook work: a Streamlit app wrapping Google's PaLM model, and a deployed generative Hangman game built on generative AI rather than a static word list. Both sit alongside his Kaggle Grandmaster competition notebooks and Google Developer Expert (ML) recognition as evidence of range from classic TensorFlow/Keras pipelines to current GenAI tooling on GCP/Vertex AI.

Wrapping a foundation model in a usable interface and getting it deployed is closer to the actual day-to-day of an ML engineer than a notebook that stops at a leaderboard metric — combined with Kaggle Grandmaster and GDE status, it shows both adversarial modeling skill and the separate skill of shipping on a cloud platform.

  • Streamlit app built on top of Google's PaLM model, not just a local script

  • Generative Hangman game deployed as a live, playable app rather than a rule-based demo

  • Paired with Kaggle Grandmaster rank and Google Developer Expert (ML) status as independent, third-party validation of the underlying skill

Data AnalysisintermediatePublished on Kaggle c. 2018 (listed as 'Updated 8 years ago')

Deep Learning for Time Series Forecasting (Kaggle notebook)

PythonKerasTensorFlowpandasscikit-learnPlotly

A Kaggle notebook that builds and compares four deep learning approaches to univariate forecasting on the Store Item Demand Forecasting data (913,000 daily store-item sales rows, 2013–2017): a Multilayer Perceptron, a 1D CNN, an LSTM and a hybrid CNN-LSTM. It walks through reframing a time series as a supervised learning problem with shifted lag features, with a 90-day forecast horizon. The notebook credits a machinelearningmastery.com mini-course as its inspiration.

It shows how to set up a forecasting problem, not just fit one model: the reframing step and a side-by-side comparison of four architectures. The vote count shows other practitioners found it worth reusing.

  • MLP vs. CNN vs. LSTM vs. CNN-LSTM on the same sales data

  • Time series reframed as supervised learning with lag features

  • 2,989 Kaggle upvotes, 114 comments

ML ModeladvancedBuilt June 2021; last updated October 2023

BERT as a Service (TFX pipeline)

TensorFlow Extended (TFX)BERTTF HubHugging FaceKerasTunerKubeflowVertex AITensorFlow Serving

An end-to-end TFX pipeline that trains and serves a BERT sentiment model on the IMDB dataset, built to demonstrate MLOps practice rather than a single notebook. The README walks through each component: ExampleGen, StatisticsGen, SchemaGen and ExampleValidator for data checks, Transform, a KerasTuner Tuner, Trainer, Evaluator and InfraValidator, ending with a model served over REST or gRPC. It ships pipeline variants for Colab (BERT from TF Hub or Hugging Face) and for GCP on Kubeflow and Vertex AI.

Most portfolio ML stops at a trained model. This one covers data validation, tuning, model evaluation against a baseline, infra validation and serving, on two orchestrators. That is the ML engineering half of the job.

  • Full TFX component chain from ingestion to REST/gRPC serving

  • Runs in Colab or on GCP via Kubeflow and Vertex AI

  • InfraValidator checks the model in real serving infra before push

Ready to build your portfolio?

Follow the ML Engineer roadmap — skills, projects, and timeline to get hired.

ML Engineer Roadmap