AI-900 Practice Question: Describe features of Natural Language Processing workloads on Azure

Answer analysis

Option-by-option breakdown

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

• ✗

  Scheduling language model inference jobs to run at off-peak hours

  Why it's wrong here

  Scheduling language model inference jobs to run at off-peak hours is an infrastructure cost-optimization pattern, not an orchestration workflow. Orchestration in Azure AI Language is a request-time routing layer that selects among skills like Custom Language Understanding (CLU) or custom question answering based on the user's intent, independent of when compute jobs execute. This option confuses deployment-time or batch processing concerns with the dynamic, real-time skill selection that orchestration provides.

• ✓

  A single entry point that routes queries to the appropriate language skill based on intent

  Why this is correct

  An orchestration workflow is a single, unified entry point—typically deployed as an Azure AI Language resource—that uses a top-level intent classifier to route each incoming query to the most appropriate underlying language skill, such as CLU for custom intents, custom question answering for knowledge bases, or LUIS for legacy apps. It centralizes skill connectivity and lets developers build a conversational application that exposes one endpoint while still leveraging multiple specialized language services. This dynamic, intent-based routing is the core purpose of orchestration, distinguishing it from scheduling, preprocessing, or data-labeling tasks.

• ✗

  Automating the data labelling workflow for custom NLP model training

  Why it's wrong here

  Automating the data labelling workflow for custom NLP model training is an ML-assisted annotation activity, part of the model development lifecycle (e.g., using active learning or pre-labelling to reduce human effort). Orchestration is a runtime serving feature that routes incoming queries across multiple already-deployed language skills from a unified endpoint; it has no role in creating or curating training datasets for those skills. This option mistakes a model-building phase for the inference-time routing architecture that orchestration provides.

• ✗

  Managing the sequence of preprocessing steps applied to text before model inference

  Why it's wrong here

  Managing the sequence of preprocessing steps applied to text before model inference (e.g., tokenization, normalization, or stemming) is an NLP engineering task that occurs within a single language processing pipeline. An orchestration workflow operates at a higher semantic level, using a top-level classifier to route the entire utterance to the appropriate language skill, rather than modifying the text content before a skill processes it. Thus, preprocessing is a lower-level component concern, not the cross-skill routing mechanism orchestration defines.

Go deeper

Related to this question