AWS Certified Generative AI Developer - Professional (AIP-C01) Passed Study Notes
--Introduction
I recently passed the AWS Certified Generative AI Developer - Professional. I took it because I wanted to be eligible for the exclusive swag at the upcoming AWS Summit.
It was a close call, but I passed! With this, I’ve now truly completed all AWS certifications.
Here’s the exam overview:
https://docs.aws.amazon.com/aws-certification/latest/ai-professional-01/ai-professional-01.html
This exam is intended for individuals who perform a GenAI developer role. The exam validates a candidate’s ability to effectively integrate foundation models (FMs) into applications and business workflows. This certification demonstrates practical knowledge about how to implement GenAI solutions into production using AWS technologies.
Study Notes
Amazon Bedrock Knowledge Bases
A managed service for building RAG. It handles the vector database and embedding model setup for you, so all you really need to do is connect your data. Supported vector DBs include Amazon S3 Vectors, Amazon Aurora PostgreSQL (pgvector), Amazon OpenSearch Serverless, Pinecone, and more
There are two APIs for invoking RAG: RetrieveAndGenerate and Retrieve. You can also apply metadata-based filters. Comes with source attribution. The following video is super helpful
Other references:
Amazon Bedrock Guardrails
You can set up guardrails that apply to both input prompts and output results. Supports topic blocking, PII masking, prompt injection defense, etc
Easy to understand:
Amazon Bedrock AgentCore
Has a bunch of components I had AI summarize this one
| Component | Role |
|---|---|
| Runtime | Secure serverless execution environment. Isolated microVMs per session, scales instantly to thousands |
| Memory | Short-term (per session) and long-term (shared) memory management. State persistence across sessions |
| Gateway | Converts existing APIs/Lambda into MCP-compatible tools, connects to external MCP servers |
| Identity | Auth and access control integrated with Cognito/Entra ID/Okta/OAuth (M2M auth) |
| Browser | Cloud-based secure browser runtime (agents can do web operations) |
| Code Interpreter | Secure code execution in a sandboxed environment |
| Observability | Agent visibility through distributed tracing, metrics, and logs |
| Policy | Centralized control over agent actions |
- Strands Agents: AWS-native agent orchestration. Organizes multiple tools into controlled flows (Strands) like “decode→correlate→normalize→route”
- AWS Agent Squad: Multi-agent system building
Easy to understand:
Amazon Bedrock Prompt Management
Version control for your prompts
Amazon Bedrock Flows
Some cool thing that lets you build flows
Amazon Bedrock Model Evaluation
You can have LLMs judge outputs using LLM-as-a-Judge
AWS AppConfig
Lets you change application parameters without deploying