AIF-C01
Deep Dive
Bedrock is AWS's central generative AI service and probably the most tested one on the exam. It gives access to foundation models from multiple providers through a unified API, without managing your own infrastructure.
Contents
Amazon Bedrock is a fully managed service that exposes foundation models through a consistent REST API. "Serverless" means AWS fully manages the hardware (GPUs), scaling, and availability — the user just sends inference requests and pays for what they consume, without provisioning or administering any instance.
What Bedrock does NOT do
Bedrock does not train foundation models from scratch — it gives access to models already trained by Amazon or third parties. To train your own model from scratch, the corresponding service is SageMaker.
SageMaker JumpStart — a complement, not a competitor
SageMaker JumpStart is a hub of pre-trained models (foundation models and others) that can be deployed with one click directly onto your own SageMaker infrastructure, with more control over the deployment environment than Bedrock. It's complementary to Bedrock: Bedrock offers serverless API access to managed foundation models; JumpStart lets you deploy models (including some from the same providers) onto SageMaker endpoints that you manage.
Invoke your first foundation model from the Amazon Bedrock playground.
+1 more...
Adjust the temperature parameter in Bedrock Playground and observe its effect on the model's response.
+1 more...
Instead of integrating separately with each AI provider's own API, Bedrock offers a single, consistent interface to invoke models from different makers: Amazon (Titan / Nova family), Anthropic (Claude), Meta (Llama), Mistral AI, among others available depending on the region.
This lets you compare models from different providers for the same use case, or switch models with minimal code changes, without managing separate credentials or infrastructure for each provider.
Bedrock Knowledge Bases is AWS's managed implementation of RAG (Retrieval-Augmented Generation): it connects your own documents to a foundation model without you having to manage a vector database or the retrieval pipeline separately. Covered in more depth in the D3 module (Foundation model applications).
Bedrock Agents lets a foundation model orchestrate concrete actions: calling APIs, executing Lambda functions, or querying external systems from natural language instructions, automatically chaining multiple reasoning steps. Covered in more depth in the D3 module.
Bedrock Guardrails is a configurable safety layer applied to both user inputs and model responses, filtering:
Harmful content
Blocks topics like violence, hate, explicit sexual content, self-harm
Denied topics
Defines specific topics the business doesn't want the model to discuss (e.g. legal or medical advice)
PII (personal information)
Detects and redacts (hides) sensitive data such as card or ID numbers
Prompt injection and jailbreaks
Detects attempts to manipulate the model into bypassing its original instructions
Key point for the exam
Guardrails filters and controls content for safety and compliance reasons — it doesn't improve the model's accuracy or factual knowledge. Don't confuse it with RAG (which does improve accuracy by adding real context).
Bedrock lets you customize existing foundation models with your own data, without needing to train a model from scratch as you would with SageMaker. Two main approaches:
Fine-tuning
Adjusts the base model's weights using your own labeled dataset (prompt-response pairs), creating a private version of the model specialized in your business's style, tone, or domain.
Continued pre-training
Continues the base model's pre-training with your own unlabeled data (for example, extensive internal documentation), broadening its domain knowledge before applying prompt engineering or fine-tuning.
The resulting customized models remain private to your AWS account — they are not shared with or used to train the base model for other customers.
Did you get this topic?
Test what you just learned
A company uses Amazon Bedrock for its public chatbot and needs to prevent the model from answering questions about medical topics and from exposing credit card numbers that users might accidentally type into the chat. Which Bedrock feature directly addresses this requirement?
Put what you learned to the test with original AIF-C01 exam questions.
Start AIF-C01 mock examUnlimited mock exams with the Pro plan
Log in to track your progress.