Foundation model integration, data management, and compliance cheat sheet
All 16 AIP-C01 flashcards for Foundation model integration, data management, and compliance on one page, grouped by exam objective. Print it or save it as a PDF for a last-minute read.
Analyze requirements and design GenAI solutions
- Why build a proof of concept before full deployment?
- To validate feasibility, quality, latency and cost with real data on Amazon Bedrock before committing to an architecture.
- What does the Generative AI Lens add to the Well-Architected Framework?
- GenAI-specific best practices and questions for design, security, reliability, cost and operations, useful as standard components across projects.
Select and configure FMs
- What is a Bedrock inference profile?
- A resource you use as the model ID that routes requests across Regions (cross-Region inference) or tracks usage for an application.
- Why keep the model ID in AppConfig?
- So model selection and provider switching are configuration changes with validation and gradual rollout, not code deployments.
- What is a circuit breaker for model calls?
- A pattern (often Step Functions plus a state store) that stops calling a failing model after repeated errors and routes to a fallback or a graceful message.
- What does LoRA do?
- Low-rank adaptation trains small adapter weights instead of all parameters, so customization is cheaper and the adapters are easy to version.
Implement data validation and processing pipelines for FM consumption
- Converse vs InvokeModel?
- Converse uses one unified message format across supported models. InvokeModel passes each provider's own request body.
- What checks belong in a data validation step before FM use?
- Schema, completeness, duplicates, encoding and PII. Tools: AWS Glue Data Quality, Lambda functions and CloudWatch metrics.
Design and implement vector store solutions
- How do metadata filters help retrieval?
- They restrict the search to documents with matching attributes (department, date, tenant) before or while scoring, improving precision and access control.
- How does a Bedrock Knowledge Base stay fresh?
- Run a data source sync (ingestion job) when the source changes. It adds, updates and removes vectors.
- Name three vector store options on AWS.
- OpenSearch Serverless or OpenSearch Service, Aurora PostgreSQL with pgvector, and Amazon S3 Vectors.
Design retrieval mechanisms for FM augmentation
- Which dimensions does Titan Text Embeddings V2 offer?
- 256, 512 or 1,024 dimensions. Smaller means cheaper storage and faster search, at some cost in fidelity.
- What does hierarchical chunking return?
- Small child chunks are matched, then the larger parent chunk is returned for context.
- What does a reranker model do in RAG?
- It rescores the retrieved candidates against the query so the most relevant passages land at the top of the context.
- Why use hybrid search?
- Keyword scoring catches exact terms (product codes, names) that embeddings blur. Vectors catch meaning. Together they miss less.
Implement prompt engineering strategies and governance for FM interactions
- What do Bedrock Prompt Flows add over single prompts?
- Visual chains with prompt, condition, Lambda and knowledge base nodes, so you get branching and pre- and post-processing.