AIP-C01 tests how to build production generative AI on AWS: choosing and integrating foundation models, retrieval and agents, safety and security, cost and performance, and evaluation. This outline lists what each domain expects, in plain terms. Tick topics off as you cover them. Your ticks are saved in this browser.
This outline follows the official AIP-C01 exam guide . Check it for the exact wording.
Foundation model integration and data 31% Everything before the user sees an answer: designing the solution, choosing models, preparing data, vector stores, retrieval and prompt management. The largest domain at 31%.
Analysing requirements and designing solutions Selecting and configuring foundation models Data validation and processing for FMs Vector store solutions Retrieval mechanisms Prompt engineering and governance Instruction frameworks: Prompt Management roles, Guardrails, response templates Context and conversation: Step Functions clarification, Comprehend intent, DynamoDB history Prompt governance: parameterised templates, approvals, S3 repositories, CloudTrail and CloudWatch Logs Prompt quality assurance and regression testing Iterative refinement: structured inputs, output formats, chain of thought, feedback loops Prompt Flows: sequential chains, conditions, reusable components, pre- and post-processing
Implementation and integration 26% Agents, tools, model deployment, enterprise integration, API patterns and developer tooling. 26% of the exam.
Agentic AI and tool integrations Model deployment strategies Enterprise integration architectures FM API integrations Application patterns and developer tools
AI safety, security and governance 20% Content safety, hallucination control, privacy, governance and responsible AI. 20% of the exam.
Input and output safety controls Input protection: Guardrails, custom Step Functions and Lambda moderation Output protection: Guardrails, moderation models, text-to-SQL for deterministic results Reducing hallucinations: grounding, confidence scoring, semantic similarity checks, JSON Schema Defence in depth: Comprehend pre-filters, model guardrails, Lambda post-processing, API Gateway filtering Threat detection: prompt injection, jailbreaks, sanitisation, classifiers, automated adversarial testing Data security and privacy controls AI governance and compliance Compliance frameworks: model cards, Glue lineage, tagging, decision logs Data source tracking: Glue Data Catalog, source attribution, CloudTrail Organisational governance aligned to policy and regulation Continuous monitoring: misuse, drift, bias drift, alerting, remediation, redaction, output policy filters Responsible AI
Operational efficiency and optimization 12% Cost, throughput, latency and observability for GenAI applications. 12% of the exam.
Cost optimisation and resource efficiency Token efficiency: estimation, context optimisation, response limits, prompt compression, pruning Cost-effective model selection: tiering, price-to-performance, efficient inference High-performance systems: batching, capacity planning, auto scaling, Provisioned Throughput Caching: semantic, fingerprint, edge, deterministic hashing and prompt caching Application performance Monitoring GenAI applications Holistic observability: metrics, tracing, FM interaction tracing, business dashboards GenAI monitoring: tokens, prompt effectiveness, hallucination rate, anomaly detection, invocation logs, cost anomalies Integrated insights: compliance, audit, user interaction and behaviour tracking Tool performance: call patterns, multi-agent tracking, baselines Vector store operations: performance, index optimisation, data quality GenAI troubleshooting: golden datasets, output diffing, reasoning path tracing
Testing, validation and troubleshooting 11% Evaluating quality, validating releases and troubleshooting GenAI applications. 11% of the exam.
Evaluation systems for GenAI Assessment beyond classic ML: relevance, factual accuracy, consistency, fluency Model evaluation: Bedrock Model Evaluations, A/B and canary tests, cost-performance analysis User-centred evaluation: feedback, ratings and annotation workflows Quality assurance: continuous evaluation, regression tests, automated quality gates Multi-perspective assessment: RAG evaluation, LLM-as-a-judge, human feedback Retrieval quality testing: relevance, context matching, latency Agent performance: task completion, tool usage, Bedrock Agent evaluations, reasoning quality Reporting to stakeholders with dashboards and comparisons Deployment validation: synthetic workflows, hallucination and drift checks, consistency checks Troubleshooting GenAI applications Content handling: context window overflow, dynamic chunking, truncation errors FM integration: error logging, request validation, response analysis Prompt problems: testing frameworks, version comparison, systematic refinement Retrieval problems: relevance, embedding quality, drift, vectorisation, chunking, search performance Prompt maintenance: template tests, Logs and X-Ray pipelines, schema validation, refinement workflows