Machine Learning Engineer - Hybrid

What matters most

• Reliable model inputs

  Build and maintain data preparation and validation tooling that provides high-quality inputs for ML and optimization models.

• Production-ready ML infrastructure

  Design infrastructure for training, evaluation and deployment, and maintain model CI/CD pipelines with automated testing and validation.

• Production model visibility

  Implement monitoring, drift detection and feedback loops to track model performance after deployment.

• Cross-team productionization

  Work with applied and data scientists to move models from experimentation to deployment, collaborate with data engineering on reliable and accessible pipelines, and contribute to shared MLOps tooling.

Likely challenges to think through

• Balancing release speed with validation

  Preparation hypothesis, not a stated XPO problem: automated checks can slow deployment if they are too broad, while weak checks can miss failures. Prepare an example of how you would choose tests and validation gates for a model release.

• Handling upstream data changes

  Preparation hypothesis, not a stated XPO problem: model input quality depends on pipelines maintained across teams. Prepare to explain how you would detect a data issue, limit its effect on models and coordinate a fix with data engineering.

• Making monitoring actionable

  Preparation hypothesis, not a stated XPO problem: drift signals may not indicate a meaningful performance decline, and excessive alerts can obscure real issues. Prepare to discuss how you would pair drift checks with performance feedback and decide when to investigate.

Topics to prepare

• Data quality for ML inputs

  The role builds preparation and validation tooling for ML and optimization models. Suggested exercise: explain how you would detect missing values, schema changes and inconsistent records before training, and what should happen when a check fails.

• Training-to-deployment lifecycle

  The posting calls for infrastructure for model training, evaluation and deployment. Suggested exercise: sketch the stages from an experimental model to a production release, identifying what each stage must produce or verify.

• ML CI/CD and release checks

  Building CI/CD pipelines with automated testing and validation is a stated responsibility. Suggested exercise: distinguish tests for data inputs, model outputs and deployment readiness, then explain which failures should block a release.

• Production monitoring and feedback

  The role implements performance monitoring, drift detection and feedback loops. Suggested exercise: describe the difference between an input-data shift and a performance decline, and how you would investigate each when outcome data arrives late.

• Batch and real-time inference

  Both serving modes appear in the preferred qualifications. Suggested exercise: compare their latency, reliability and monitoring considerations. Treat this as preparation for a preferred area, not a minimum requirement.

Skills to revise

• Build a Python and SQL validation workflow

  Python and SQL proficiency and data-quality tooling are minimum qualifications. Suggested exercise: write a SQL query that identifies duplicate or missing records, then a Python check that validates the resulting dataset and reports failures.

• Automate ML pipeline checks

  The role builds ML CI/CD pipelines with automated tests and validation. Suggested exercise: create a small test sequence for data preparation and model evaluation, including a deliberately failing input and a clear release-blocking result.

• Trace a cloud ML workflow

  Experience with a cloud data or ML platform, such as AWS, GCP or BigQuery, is a minimum qualification. Suggested exercise: using one platform you know, walk through how data would be accessed, validated and made available for training and evaluation; do not assume XPO uses that platform.

• Work through a monitoring response

  Production monitoring, drift detection and feedback tooling are stated responsibilities. Suggested exercise: draft a simple monitoring plan with a metric, a drift signal, an investigation step and a route for incorporating observed outcomes.

• Practice a cross-team handoff

  The posting requires collaboration with science and data engineering teams to productionize models and improve pipeline reliability and access. Suggested exercise: prepare a concise handoff explaining the model’s input contract, validation failures, deployment checks and questions each team must resolve.

• Prepare a data-pipeline walkthrough

  Select a project, if available, and make a one-page walkthrough showing how Python and SQL handled data preparation, validation, and quality checks for ML inputs. Note any cloud platform used; these are stated qualifications.

• Sketch a productionization design

  Draw a proposed path from model experimentation through training, evaluation, and deployment. Label the data inputs, validation gates, and automated tests; present it as a design exercise, not an XPO architecture.

• Explain an ML CI/CD decision

  Prepare a short example or proposed pipeline showing what would trigger a model build, what tests would run, and what would block deployment. ML CI/CD is part of the advertised work.

• Plan production monitoring

  Create a brief monitoring outline covering model performance, drift detection, and a feedback loop. If discussing batch or real-time inference, identify it as preferred experience rather than a minimum requirement.

• Document a cross-team collaboration example

  Prepare a concise account of how data science or applied science and data engineering partners resolved a model deployment, data reliability, or data access issue. Specify each team’s contribution and the outcome.

• Bring targeted questions for XPO

  Write questions about the current path from experimentation to production, ownership of data-quality checks, and how the AI/ML organization shares MLOps tooling. These ask about responsibilities in the posting without assuming XPO’s processes.