AWS AI Practitioner Practice Test

Inference is the act of running an input through a trained model to obtain an output. Training changes model parameters, but inference uses the trained model as it already exists.

AWS documentation says Region selection is one of the main design choices for meeting legal or regulatory requirements. If the workload must keep data in the EU, the resources that store and process that data should be placed in an appropriate EU Region.

Sensitive information filters in Guardrails are specifically meant to detect, block, or mask PII and regex-based patterns in prompts and responses. Model evaluation measures quality, but it is not the enforcement control for hiding private data at runtime.

Bedrock model evaluation jobs are designed for quality and responsibility checks before rollout. They let teams assess model behavior in a repeatable, managed way rather than relying on ad hoc spot checks.

The design guide notes that placing the main question or instruction at the end of the prompt can improve results. That placement helps the model focus on the exact task it should complete next.

AWS documents Model Monitor for both real-time endpoints and batch workflows. The goal is post-deployment quality monitoring, not just observing one narrow inference style.

AWS KMS supports VPC endpoint policies that can constrain which principals, API calls, and keys are reachable through the endpoint. That is the endpoint-layer control mentioned in the KMS IAM and security documentation.

Bedrock evaluations are for quality and safety testing of models and RAG resources, whereas Audit Manager is for gathering and organizing assessment evidence. They complement each other, but they solve different lifecycle problems.

Retrieve is correct because Bedrock Knowledge Bases provides that operation when you want only the relevant chunks returned from the knowledge base. This matters because some architectures decouple retrieval from generation so each step can be tuned separately.

Grounding means supplying authoritative context so the model can answer from that material rather than from unsupported guesses. AWS positions grounding as a core trust pattern for enterprise generative AI.