AI-900 Practice Question: Describe Artificial Intelligence workloads and considerations

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Operational procedures for Microsoft 365 mail system administration

  Why it's wrong here

  Mail system administration for Microsoft 365 is IT operations focused on messaging infrastructure like Exchange Online, mail flow, spam filtering, and mailbox management—completely different from machine learning. MLOps applies DevOps automation to the ML lifecycle, including model training, evaluation, deployment, and monitoring. The toolchains and objectives differ entirely: MLOps deals with data and model pipelines, whereas M365 administration deals with email services, licensing, and compliance.

• ✓

  Applying DevOps practices (automation, CI/CD, monitoring) to the machine learning lifecycle

  Why this is correct

  MLOps is the application of DevOps principles—automation, continuous integration/continuous delivery (CI/CD), and monitoring—specifically to the machine learning lifecycle. In practice, this means automating steps from data preparation and feature engineering through model training, validation, and deployment, with versioning of data, code, and models. Monitoring in MLOps tracks model performance and data drift, triggering retraining pipelines automatically when needed, which enables consistent, reliable, and frequent model updates at scale without manual intervention.

• ✗

  A certification program for ML engineers working with Azure

  Why it's wrong here

  MLOps is not a certification program; it is an engineering and operational discipline. While Azure offers professional certifications (such as AI-900, DP-100, or AZ-104) to validate skills, those are learning credentials issued by Microsoft, not practices that operationalize ML models. The purpose of MLOps is to bridge the gap between model development and production operations using automated pipelines, version control, and monitoring, rather than to serve as an educational credential or exam track.

• ✗

  The process of optimising ML model inference speed for production deployment

  Why it's wrong here

  While MLOps does cover deployment concerns, optimizing inference speed (e.g., quantization, pruning, or using ONNX Runtime) is only a small facet of the overall discipline. MLOps encompasses the end-to-end lifecycle: automating data validation, experimenting, training, model registry management, CI/CD rollout, and post-deployment monitoring for concept drift or data drift. Inference optimization is one performance-tuning task within that lifecycle, not the definition of MLOps itself.

Go deeper

Related to this question