The correct option is Vertex AI Feature Store.

This service is designed to centralize feature computation, storage, and serving so teams can reuse a single definition across multiple models. It provides an offline store for training and an online store for low latency prediction which ensures the recommendation and fraud models read the same values and schema for the shared customer_order_rate_60d feature. It also supports consistent ingestion, versioning, and access control which reduces training and serving skew and promotes reliable reuse across projects.

Vertex AI Model Monitoring focuses on monitoring model performance and data drift after deployment. It does not compute, store, or serve reusable features for multiple models.

Vertex AI Pipelines orchestrates machine learning workflows and automates steps such as data preparation and training. It is not a purpose built repository for storing and serving features for online and offline use.

Vertex AI Matching Engine serves vector similarity search for embeddings which is useful for semantic retrieval and nearest neighbor lookups. It is not a managed feature repository for tabular or time based features shared across models.

Full AWS Practitioner Certification Question