NCA-AIIO - Essential AI Knowledge - Section 1.7

Describe software components related to the lifecycle of AI development and deployment.

Describe the software tools involved at each stage of the AI lifecycle, from data preparation and model training with frameworks like PyTorch, through experiment tracking and MLOps pipelines, to model serving at inference time. Recognise how NVIDIA tools such as TensorRT and Triton Inference Server fit into deployment and optimisation stages.

Practice question for this objective

Free sampleEssential AI Knowledgemedium

A data science team has finished cleaning and versioning their training dataset and wants to begin iterative model experimentation. Which lifecycle stage immediately follows data preparation, and what is the primary software concern at that stage?

  • AModel registry - storing approved model artefacts for controlled promotion to production
  • BInference serving - deploying a model endpoint so applications can request predictions in real time
  • CTraining framework - orchestrating compute, defining model architecture, and running gradient-based optimisation on the prepared data Correct
  • DMonitoring - tracking data drift and model performance degradation against production baselines
Identify the software stage that directly follows data preparation in the AI development lifecycle and its core function. The AI development lifecycle progresses through data preparation, training, experiment tracking and evaluation, model registry, inference serving, and then ongoing monitoring. A training framework such as PyTorch, TensorFlow, or JAX is the primary software component at the training stage: it defines the model graph, manages GPU/multi-node compute allocation, and executes forward and backward passes over the prepared dataset. Entering inference serving or monitoring before training completes skips critical lifecycle stages.

Why A is wrong: A model registry manages promotion of validated models, not the iterative experimentation that comes directly after data preparation. Registries operate after training and evaluation are complete.

Why B is wrong: Inference serving comes after a model is trained and registered. Running a serving layer on unprepared, untrained weights produces no useful predictions and is the wrong stage.

Why C is correct: After data is prepared and versioned the next stage is model training, where a framework such as PyTorch or TensorFlow defines the network, manages distributed compute, and executes the optimisation loop.

Why D is wrong: Monitoring is a post-deployment concern that tracks live model behaviour. There is no production model to monitor immediately after data preparation, so this stage does not follow data prep.

See more NCA-AIIO practice questions, answers explained.

More in this domain

Back to all Essential AI Knowledge objectives, or the NCA-AIIO cert hub.

Examworthy is not affiliated with or endorsed by NVIDIA. Original, blueprint-aligned practice material only.