The correct option is Vertex AI Feature Store.

This service is designed to manage the full lifecycle of machine learning features. It provides an offline store for historical feature computation and an online store for low latency serving, which fits the need to update and fetch fresh features within about 20 seconds of user activity. It also offers feature monitoring to detect data drift and quality issues so the team can observe behavior shifts and keep recommendations accurate over time. It integrates with common data pipelines which makes continuous refresh and governance of features manageable at scale.

Vertex AI Agent Builder focuses on building and deploying conversational and agentic experiences and it does not provide a managed repository for feature definitions, online feature serving, or feature drift monitoring.

Prompt Engineering is a technique for crafting inputs for generative models and it is not a managed Google Cloud tool for storing, serving, or monitoring feature data.

Reinforcement Learning is a training paradigm rather than a Google Cloud service and it does not manage feature storage, low latency feature serving, or feature monitoring in production.

Full AWS Practitioner Certification Question