Bernardo Galvão

MLOps Engineer Remote

Machine Learning Engineer with AWS and Kubernetes

• Created a Kubernetes CronJob connected to a Slack bot that switches off KubeFlow notebooks, therefore cutting AWS costs by releasing notebook-bound EC2 resources.
• Developed Argo Workflow for monitoring of models in production, using NannyML, Kubernetes, AWS SQS and S3 Event notifications, allowing for observability of performance and feature drift.
• Enhanced data processing workflows in AirFlow and added unit tests to AirFlow DAGs.
• Deployed and managed Grafana, Prometheus and Loki for general monitoring of K8s resources.
• Added model pre-deployment tests in CI/CD pipelines to ensure model, training and data consistency.

MLOps Engineer (Contractor) Remote

• Developed Docker Swarm stacks for deployment of MLOps services, including a model registry (MLFlow); a reverse‐proxy (Traefik) with an authentication service; a workflow orchestrator (Prefect).
• Designed an on‐prem MLOps platform to provide end‐to‐end services for frictionless training, validation, deployment and monitoring of ML models based on Open‐Source software, including a CI/CD component for ML.
• Designed a workflow template using DVC for Data Scientists to confidently develop datasets and models with unit testing integrating a CI/CD pipeline written in GitLab CI/CD specification.
• Wrote tests and GitLab CI pipelines for deployment of thoroughly tested models on their data and performance. Wrote Gitlab CI pipelines for automated deployment of Docker Stacks.
• Optimized performance of data queries and transformations using optimized memory formats Apache Arrow.

Research Fellow Lisbon, Portugal

• Deployed MLFlow server for experiment tracking and experiment data collection.
• Conducted research on performance feature selection methods applied to radiomics for clinical decision‐making.
• Implemented Target Shuffling robustness check and Nested Cross‐Validation procedure for use by other members in the team.

Data Scientist

• Implemented an object detection viewer and trained a marine‐species object detector via transfer learning.
• Automated dataset processing, model training and Tensorflow model format conversion to TensorFlowJS and TensorFlow Lite using Docker, Python and Bash.
• Provided a user‐friendly CLI for pulling, configuring and training models from TensorFlow’s Object Detection API.
• Implemented process for cooperative collection, annotation and augmentation of training images.
• Analyzed social media data activity levels per location, including profiling points of interest on Madeira island via topic modelling of TripAdvisor reviews with Latent Drichlet Allocation.
• Performed data wrangling and analysis to assess performance of low‐cost air quality sensors.

Software Engineering Intern Lisbon, Portugal

• Developed a raycasting prototype in order to produce an attention heatmap on a 3D object using NumPy and VTK in Python.