The correct approaches are Bedrock supports prompt-based customization without changing model weights and Bedrock offers managed fine-tuning on labeled, domain-specific examples without operating infrastructure.

Bedrock supports prompt-based customization without changing model weights lets the manufacturer use instruction design and few shot examples to teach the model specialized engineering jargon and acronyms while avoiding any parameter updates. This approach is quick to try and it is useful for steering behavior when you want low cost experimentation or task specific formatting.

Bedrock offers managed fine-tuning on labeled, domain-specific examples without operating infrastructure lets you create durable domain adaptation by providing labeled examples and letting Bedrock perform the training and parameter updates. This approach is appropriate when you need consistent, long lived improvements for interpreting technical terms and you want AWS to handle the training infrastructure.

Bedrock lets you train a foundation model from scratch using only raw engineering documents is incorrect because Bedrock provides access to pretrained foundation models and it does not support training a foundation model from scratch on your own raw corpus.

Knowledge Bases for Amazon Bedrock automatically retrain model parameters when you ingest documents is incorrect because Knowledge Bases supply retrieval augmented context to prompts and they do not change the model weights when you add documents.

Bedrock supports model deployment only through Amazon SageMaker pipelines is incorrect because Bedrock exposes its own inference APIs so you can deploy and call models without requiring SageMaker pipelines.

Full AWS Practitioner Certification Question