AWS Generative AI Developer Study Topics and Tips (AIP-C01)
64 key facts from the official AWS documentation, grouped into 8 study topics. Each note links to the AWS page it comes from, so you can verify it and go deeper. In the Android app you get one every 3 hours when you choose this exam.
RAG data: chunking and embeddings
Fixed-size, hierarchical, semantic and no chunking in Bedrock Knowledge Bases, plus Titan Text Embeddings V2 dimensions and binary vectors.
7 notes
RAG retrieval: hybrid search, reranking and filters
How Bedrock Knowledge Bases retrieve: hybrid vs semantic search, reranking, metadata filters, query decomposition and Retrieve vs RetrieveAndGenerate.
7 notes
Bedrock Guardrails: filters, PII and grounding
Guardrail policy types, content filters, denied topics, PII block vs mask, contextual grounding thresholds, detect mode and Automated Reasoning.
13 notes
- Guardrail policy types
- Content filters and prompt attacks
- Denied topics vs word filters
- Block or mask PII
- Masking does not protect your logs
- Tool use is not filtered for PII
- Contextual grounding check
- Grounding thresholds
- Grounding limits and tiers
- Grounding checks and streaming
- Detect mode
- Automated Reasoning checks
- Guardrail trace shows the original PII
Governance: enforcing guardrails and data privacy
Enforce guardrails with the bedrock:GuardrailIdentifier IAM condition, understand Bedrock data privacy, and allow cross-Region inference in SCPs.
5 notes
Capacity: throughput, cross-Region inference and routing
Provisioned Throughput, geographic vs global inference profiles, batch inference and intelligent prompt routing for the AIP-C01 exam.
7 notes
Prompt caching and token costs
How Bedrock prompt caching works: static prefix first, checkpoints, TTL, cache misses and how to count total input tokens with caching on.
3 notes
Orchestration: Step Functions, streaming and retries
Step Functions limits and callbacks, Lambda response streaming, SDK retry modes, the Converse API, Prompt Management and AWS AppConfig for AIP-C01.
15 notes
- Streaming APIs
- Lambda response streaming
- Response streaming details
- 256 KiB payload limit
- Standard vs Express
- Human approval with a callback token
- Heartbeat on waiting tasks
- Standard retry mode
- Adaptive retry mode
- Safeguard agent workflows
- MCP servers on AWS
- One API for every chat model
- Reusable prompts, variants and versions
- Change behavior without redeploying
- AppConfig Agent
Monitoring, logging and evaluation
Model invocation logging destinations and limits, per-caller token usage, Bedrock evaluation job types and RAG evaluation datasets for AIP-C01.
7 notes
Looking for security? See the Security Specialty study topics.
Test what you just read
A note sticks when you use it. Answer practice questions on these topics.