10 Amazon Bedrock Concepts to Know for AIP-C01

The core building blocks

Bedrock exposes generative AI through a single API surface, so the exam tests whether you know which capability solves which problem rather than low-level model internals.

• Foundation models - the pre-trained models you invoke through a unified API; know that different families suit different tasks (text, chat, embeddings, images).
• Prompt engineering - structuring instructions, context, and examples to steer a model; the cheapest lever before you reach for fine-tuning.
• Embeddings - vector representations of text that power semantic search and retrieval.
• Knowledge Bases - managed retrieval-augmented generation (RAG) that grounds answers in your own documents.
• Guardrails - configurable policies that filter harmful content and block disallowed topics or sensitive data.

Going beyond a single call

Real applications rarely stop at one model invocation. The exam expects you to recognize the features that add orchestration, customization, and oversight.

• Agents - orchestrate multi-step tasks, calling tools and APIs to complete a goal.
• Fine-tuning and continued pre-training - adapt a foundation model to your domain when prompting alone is not enough.
• Model evaluation - compare models on quality, cost, and latency before committing.
• Provisioned throughput - reserve capacity for predictable, high-volume workloads.
• Observability and cost controls - log invocations and monitor spend so a generative-AI feature stays affordable in production.

Tying it together

A common exam pattern is a chatbot that must answer from company data without hallucinating: the intended answer combines a foundation model, a Knowledge Base for RAG, and a Guardrail for safety. Practice recognizing which building blocks a scenario is really asking for.