GCP Generative AI Leader Practice Question 62: ML Lifecycle Data Preparation
Google Cloud Generative AI Leader · Test 7 · question 2 of 10
A retailer has ingested three years of raw product reviews into Cloud Storage for a gen AI project. Before the data can be used to fine-tune a model, the team must clean, deduplicate, mask personal information, and transform the reviews into a consistent format at scale.
Which stage of the ML lifecycle is the team now entering, and which Google Cloud tool fits it?
Show answer and explanation
Answer:
Once raw data has already landed in storage, the next lifecycle stage is data preparation: cleaning records, removing duplicates, masking sensitive fields, and reshaping the data into a consistent format the model can consume. Dataflow is Google Cloud's managed service for exactly that kind of large-scale, transformation-heavy pipeline, running as either a batch job or a streaming job.
- A is wrong because monitoring happens after a model is trained and deployed, not while raw data is still being cleaned. - B is wrong because ingestion is the earlier stage of first bringing data into the pipeline; the reviews here are already ingested and sitting in Cloud Storage. - C is wrong because evaluation compares a trained model's outputs against metrics, which cannot happen before the training data itself is even prepared.
Reference: Dataflow overview