Hard Difficulty Questions

IAM role permissions

Why wrong: IAM issues typically cause 403 Forbidden.

Input JSON schema validation

Why wrong: Schema issues cause 400 Bad Request.

Semantic chunking based on paragraph breaks.

Why wrong: Semantic chunking is effective but often less precise for highly technical documents than hierarchical approaches.

Fixed-size chunking with 100 token overlap.

Why wrong: Fixed-size often leads to loss of context in technical legal documents.

Default chunking strategy.

Why wrong: The default strategy is often too simplistic for dense domain-specific content.

AWS Glue DataBrew

Why wrong: Glue DataBrew is for data preparation, not model fine-tuning.

Amazon SageMaker JumpStart

Why wrong: While JumpStart helps deploy models, Bedrock Custom Models is the specific pathway for fine-tuning within the Bedrock service family.

Amazon Kendra

Why wrong: Kendra is a search service, not a platform for model training.

Publicly accessible S3 buckets

Why wrong: Public buckets are a security violation.

Temporary local storage on the training instance

Why wrong: Training instances are ephemeral; data should be stored in durable, compliant storage.

DynamoDB tables

Why wrong: Fine-tuning typically requires file-based input (S3), not NoSQL table formats.

Set the IAM role to 'AdministratorAccess'

Why wrong: IAM does not govern VPC network connectivity.

Update the JumpStart model Dockerfile

Why wrong: You should not modify the provided JumpStart container images.

Enable AWS PrivateLink on the S3 bucket

Why wrong: S3 access is distinct from container VPC access.

Use AWS Config to monitor bucket changes.

Why wrong: Config is for compliance and tracking, not triggering data workflows.

Set the knowledge base to 'Auto-Refresh' mode.

Why wrong: There is no 'Auto-Refresh' mode; synchronization must be triggered.

Enable S3 Versioning.

Why wrong: Versioning is for data safety, not index triggering.

The number of concurrent IAM users.

Why wrong: IAM is not a cost driver for Bedrock.

The frequency of S3 bucket scans.

Why wrong: S3 scanning is not a direct cost factor.

Maintaining the physical security of the data center.

Why wrong: AWS responsibility.

Patching the underlying foundation model code.

Why wrong: AWS responsibility.

Exclude all logging to protect privacy

Why wrong: Logging is required for compliance.

Allow users to train the model directly

Why wrong: This is a security and governance risk.

AWS CloudTrail logs

Why wrong: CloudTrail logs activity but does not perform real-time data filtering.

AWS WAF on the API Gateway

Why wrong: WAF filters web traffic but lacks the LLM-specific logic to redact PII from natural language generation.

IAM Policy evaluation for Bedrock

Why wrong: IAM policies control access, not content sanitization.

AWS CodePipeline.

Why wrong: Used for CI/CD.

AWS AppConfig.

Why wrong: Used for configuration management.

AWS Trusted Advisor.

Why wrong: Focuses on cost and best practices, not audit logs.

Reduce the index replication factor

Why wrong: This affects availability, not ingestion performance.

Disable vector search

Why wrong: This defeats the purpose of the application.

Increase the number of replica shards

Why wrong: Replica shards add overhead to index writes.

S3 bucket storage size.

Why wrong: Not a model performance metric.

Number of active IAM roles.

Why wrong: Not a model performance metric.

By prepending the metadata to the prompt.

Why wrong: This is inefficient and does not work for filtering.

It is done automatically by the model.

Why wrong: The developer must define the metadata ingestion.

It is not possible to use metadata in Bedrock RAG.

Why wrong: It is a supported and common feature.

Using AWS Glue for data preparation only

Why wrong: Glue is for ETL, it does not secure the transit of data to Bedrock.

Deploying the model in an on-premises data center

Why wrong: Bedrock is a managed service; it cannot be deployed on-premises.

Encrypting the dataset with AWS CloudHSM

Why wrong: Encryption protects the data at rest but does not establish a private path for training.

Disable all CloudWatch logging to prevent data leakage

Why wrong: Logging is usually required for compliance audits; disabling it hinders tracking.

Store all prompts in plain text in S3 for auditing

Why wrong: Storing prompts in plain text is a security risk if not properly managed.

Increasing the instance count

Why wrong: Instance count doesn't directly solve cold starts for serverless.

Changing the container image

Why wrong: Container size affects cold starts, but Provisioned Concurrency is the direct feature.

Using a smaller model

Why wrong: Model size is not the only factor in cold starts.

Model's popularity

Why wrong: Performance metrics matter more than popularity.

Model training cost

Why wrong: Embeddings are used as-is, not trained.

Amazon SageMaker Model Monitor.

Why wrong: SageMaker Model Monitor is for tabular/ML models, not specifically for LLM text toxicity evaluation.

A

Why wrong: A

AWS Glue to sanitize data.

Why wrong: Glue is for data transformation, not model output evaluation.

AWS Lambda to call the model and manually inspect outputs.

Why wrong: Manual inspection is not automated validation.

IAMUserLoginCount

Why wrong: This is irrelevant to endpoint load.

S3RequestLatency

Why wrong: This is for storage access, not inference.