Generative AI Developer Professional

Who Should Take This

It is intended for software engineers, data scientists, or cloud architects with at least three years of experience designing AI/ML systems, who need to certify expertise in deploying and managing production‑ready generative AI workloads on AWS. They aim to demonstrate mastery of security, compliance, performance tuning, and troubleshooting to meet enterprise governance and reliability standards.

What's Covered

1 Select and integrate foundation models via Amazon Bedrock, manage training and inference data, and ensure regulatory compliance for GenAI workloads.

2 Implement RAG architectures, knowledge bases, agentic workflows, and integrate foundation models into applications and business processes.

3 Implement guardrails, content filtering, IAM controls, and governance frameworks for responsible generative AI deployments.

4 Optimize latency, throughput, and cost of GenAI applications, and manage model selection and inference configuration.

5 Evaluate and validate GenAI application outputs, implement testing strategies, and troubleshoot model performance and integration issues.

Exam Structure

Question Types

Multiple Choice
Multiple Response
Matching
Ordering

Scoring Method

Scaled scoring from 100 to 1000, minimum passing score of 750

Delivery Method

Pearson VUE testing center or online proctored

Recertification

Recertify every 3 years by passing the current exam or earning a higher-level AWS certification.

What's Included in AccelaStudy® AI

Adaptive Knowledge Graph

Practice Questions

Lesson Modules

Console Simulator Labs

Exam Tips & Strategy

20 Activity Formats

Course Outline

75 learning goals

1 Domain 1: Solution Design for Business Applications

5 topics

Foundation model selection and configuration

Identify foundation model families available through Amazon Bedrock (Anthropic Claude, Amazon Titan, Meta Llama, AI21 Jurassic, Cohere, Stability AI) and explain their modality support, context window sizes, embedding dimensionality options, and primary use cases.
Configure Bedrock model access requests, inference parameters (temperature, top-p, top-k, max tokens, stop sequences), and model-specific payload formats for text generation, embedding, and image generation workloads.
Analyze foundation model selection tradeoffs across inference cost, context window utilization, latency profiles, safety behavior, and customization pathways to match business application requirements.
Define model selection strategy that aligns modality requirements, quality expectations, regulatory constraints, and total cost of ownership for enterprise generative AI adoption.

Prompt engineering strategies

Identify prompt engineering techniques (zero-shot, few-shot, chain-of-thought, ReAct, tree-of-thought) and explain when each technique improves response quality for different task types.
Implement prompt templates with system instructions, role definitions, output format constraints, and few-shot examples that produce consistent, policy-aligned model responses across Claude and Titan models.
Implement prompt chaining and decomposition patterns that break complex tasks into sequential model calls with intermediate validation and context assembly between steps.
Analyze prompt effectiveness using evaluation metrics (relevance, faithfulness, coherence, toxicity scores) and iteratively refine prompts based on production response quality data.

Data pipelines and ingestion for FM consumption

Identify AWS data ingestion services (S3, Glue, Lambda, Kinesis Data Firehose) and explain how each supports document processing, chunking, and metadata enrichment pipelines for generative AI workloads.
Implement document chunking strategies (fixed-size, sentence-based, semantic, recursive character splitting) with overlap windows and metadata preservation for optimal embedding and retrieval quality.
Implement data validation, transformation, and quality assurance pipelines using Lambda, Step Functions, and S3 event notifications to ensure clean, policy-compliant input for model inference and knowledge base sync.
Analyze data pipeline design tradeoffs balancing freshness, chunk granularity, embedding cost, lineage tracking, and retrieval quality for production RAG workloads at scale.

Vector store design and embedding strategies

Identify AWS vector store options (OpenSearch Serverless, Amazon Aurora pgvector, Amazon Neptune, Pinecone on AWS) and explain their indexing algorithms, distance metrics, and filtering capabilities.
Implement embedding generation pipelines using Amazon Titan Embeddings and Cohere Embed via Bedrock, including batch embedding, dimensionality selection, and normalization for vector store ingestion.
Implement OpenSearch Serverless vector search collections with k-NN index configuration, HNSW parameters, metadata filtering, and hybrid search combining keyword and semantic retrieval.
Analyze vector store design decisions across embedding dimensionality, index type (HNSW vs IVF), recall-latency tradeoffs, metadata filter selectivity, and cost-performance scalability.

Retrieval-augmented generation architecture

Identify RAG architecture components (query encoding, retrieval, reranking, context assembly, augmented generation) and explain how Bedrock Knowledge Bases orchestrate the end-to-end retrieval-generation pipeline.
Implement Bedrock Knowledge Bases with S3 data sources, automatic chunking, embedding model selection, and OpenSearch Serverless vector store integration for managed RAG pipelines.
Implement custom RAG workflows with query reformulation, hypothetical document embeddings (HyDE), multi-query retrieval, reranking, contextual compression, and Amazon Kendra integration for improved retrieval relevance.
Define RAG architecture strategy balancing retrieval relevance, hallucination risk, latency budgets, context window utilization, and data freshness requirements for enterprise production workloads.

2 Domain 2: Development of Generative AI Applications

6 topics

Bedrock API integration and inference patterns

Identify Bedrock API operations (InvokeModel, InvokeModelWithResponseStream, Converse, ConverseStream) and explain their request/response formats, streaming behavior, and model-specific payload differences.
Implement Bedrock inference calls using the AWS SDK with proper credential resolution, retry configuration, streaming response handling, and error management for throttling and model timeout scenarios.
Implement the Bedrock Converse API for multi-turn conversations with message history management, tool use definitions, and cross-model compatibility for portable application code.
Analyze Bedrock API integration behavior to optimize reliability, token usage, latency distribution, and fallback paths across model providers and regions.

Agentic AI solutions with Bedrock Agents

Identify Bedrock Agents components (agent instructions, action groups, knowledge bases, guardrails attachment) and explain how the agent orchestration loop processes user requests through reasoning and tool invocation.
Implement Bedrock Agents with action groups backed by Lambda functions, OpenAPI schema definitions, and return-of-control patterns for multi-step task completion workflows.
Implement agent memory and session management with conversation history, session attributes, and knowledge base integration for context-aware multi-turn agent interactions.
Analyze agent architecture tradeoffs across autonomy scope, tool invocation boundaries, error recovery patterns, hallucination containment, and operational reliability requirements.
Define agentic workflow strategy determining when to use single agents, multi-agent collaboration, or orchestrated pipelines based on task complexity, safety constraints, and latency budgets.

Model customization: fine-tuning and continued pre-training

Identify Bedrock model customization methods (continued pre-training, fine-tuning, RLHF) and SageMaker fine-tuning approaches (JumpStart, LoRA/QLoRA, custom containers) and explain when each is appropriate based on data availability and quality goals.
Implement Bedrock fine-tuning jobs with JSONL training data preparation, hyperparameter configuration (epochs, batch size, learning rate), provisioned model throughput allocation, and custom model deployment.
Implement SageMaker-based fine-tuning for foundation models using JumpStart model selection, LoRA/QLoRA parameter-efficient methods, distributed training on GPU instances, and model artifact deployment to endpoints.
Analyze fine-tuning outcomes using validation loss curves, benchmark comparisons, and A/B evaluation to determine whether customization achieves sufficient quality improvement over prompt engineering alone.

LangChain and orchestration framework integration

Implement LangChain applications with Bedrock LLM and ChatBedrock wrappers, prompt templates, output parsers, retrieval chains, and conversational memory for structured generative AI workflows on AWS.
Analyze orchestration framework tradeoffs between LangChain, Bedrock Agents, and custom Step Functions-based pipelines based on flexibility, vendor lock-in, observability, and production reliability requirements.

Model deployment and endpoint management

Identify Bedrock deployment options (on-demand throughput, provisioned throughput, custom model hosting) and SageMaker endpoint types (real-time, serverless, asynchronous) for foundation model serving.
Implement SageMaker endpoints for custom or fine-tuned models with instance selection, auto-scaling policies, model variants, and A/B traffic routing for production model serving.
Define deployment strategy selecting between Bedrock managed hosting and SageMaker self-managed endpoints based on customization needs, latency requirements, cost constraints, and operational overhead tolerance.

Enterprise integration and application architecture

Implement serverless generative AI applications using Lambda, API Gateway, and Bedrock with proper timeout configuration, payload size management, and asynchronous invocation patterns for long-running inference.
Implement event-driven generative AI pipelines using S3 event notifications, SQS queues, Step Functions orchestration, and Lambda to process documents, generate summaries, and enrich data stores asynchronously.
Implement conversational AI interfaces using Amazon Lex with Bedrock integration, intent-based routing with generative fallback, and Lambda fulfillment hooks for hybrid structured and open-domain user interactions.
Define integration architecture strategy supporting multi-team ownership, secure service boundaries, API versioning, and long-term platform evolvability for enterprise generative AI systems.

3 Domain 3: Security, Compliance, and Governance

5 topics

Bedrock Guardrails and content safety

Identify Bedrock Guardrails components (content filters, denied topics, word filters, sensitive information filters, contextual grounding checks) and explain how each control type blocks or modifies unsafe model interactions.
Implement Bedrock Guardrails with configurable content filter strengths, denied topic definitions, PII detection and redaction policies, and custom word filters applied to both input prompts and model responses.
Implement contextual grounding checks in Guardrails to detect hallucinated responses by validating model output against retrieved source documents and filtering responses that lack grounding evidence.
Analyze guardrail effectiveness using blocked request rates, false-positive impact on legitimate queries, adversarial bypass testing, and guardrail version comparison to optimize safety without degrading utility.

IAM and access control for generative AI services

Identify IAM policy patterns for Bedrock access control and explain how resource-based policies, model access permissions, and service-linked roles govern model invocation, customization, and knowledge base operations.
Implement least-privilege IAM policies for Bedrock with model-level invocation permissions, knowledge base access controls, guardrail management rights, VPC endpoint policies, and cross-account model sharing using resource policies.
Analyze access control configurations to identify over-permissioned roles, missing model-level restrictions, and network exposure gaps across generative AI service integrations.

Data security and privacy for generative AI

Identify data handling practices for Bedrock (model invocation logging opt-in, data not used for training, encryption at rest and in transit) and explain the AWS shared responsibility model for generative AI workloads.
Implement encryption controls for generative AI data using KMS customer managed keys for Bedrock model customization data, knowledge base storage, S3 training datasets, and CloudWatch log groups.
Implement PII detection, masking, and redaction pipelines using Amazon Comprehend, Bedrock Guardrails sensitive information filters, and Macie scanning to prevent sensitive data leakage through model interactions.
Define data governance strategy for generative AI enforcing classification, retention, access controls, training data provenance, and cross-border compliance for regulated industries.

AI governance, compliance, and audit

Implement model invocation logging using CloudTrail and Bedrock model invocation logging to capture request/response metadata, token counts, guardrail actions, and model version records for audit and compliance evidence.
Define compliance strategy mapping AI regulatory requirements (EU AI Act, NIST AI RMF) to technical controls, audit mechanisms, and organizational governance processes for enterprise generative AI systems.

Responsible AI and ethical practices

Identify responsible AI principles (fairness, transparency, explainability, safety, privacy, accountability) and explain how AWS services and features support each principle in generative AI applications.
Implement responsible AI practices including bias detection with SageMaker Clarify, source attribution in RAG responses, model card documentation, confidence indicators, and user-facing transparency disclosures for AI-generated content.
Analyze responsible AI outcomes to establish remediation strategies for detected bias, misuse patterns, trust failures, and evolving ethical standards across production generative AI applications.

4 Domain 4: Performance Optimization

3 topics

Inference cost optimization

Identify Bedrock pricing dimensions (input tokens, output tokens, provisioned throughput, model customization) and explain cost drivers across on-demand, batch, and provisioned inference modes.
Implement cost optimization techniques including semantic caching, prompt compression, model routing (selecting cheaper models for simpler queries), token budget policies, and batch inference for non-real-time workloads.
Analyze inference cost patterns and define resource-efficiency strategy balancing service-level objectives, response quality targets, and total operating cost constraints across model tiers and deployment modes.

Latency and throughput optimization

Implement latency optimization techniques including response streaming, provisioned throughput for consistent performance, prompt caching, parallel retrieval, and context window size management.
Implement throughput scaling using Bedrock provisioned throughput, SageMaker endpoint auto-scaling, request queuing with SQS, and cross-region inference distribution for high-concurrency workloads.
Analyze end-to-end latency bottlenecks across retrieval, embedding, inference, and post-processing stages using tracing data to prioritize performance remediation with maximum user experience impact.
Define performance architecture strategy balancing latency SLAs, throughput capacity planning, regional deployment topology, and provisioned vs. on-demand cost models for production workloads.

RAG and retrieval performance tuning

Implement retrieval performance optimization by tuning chunk sizes, overlap ratios, embedding model selection, k-NN search parameters, and hybrid search weighting for improved recall and precision.
Analyze retrieval quality using relevance metrics (NDCG, MRR, recall@k), answer faithfulness scores, and A/B testing to identify and resolve retrieval degradation patterns in production RAG systems.

5 Domain 5: Monitoring and Troubleshooting

3 topics

Observability for generative AI applications

Identify AWS observability services for generative AI (CloudWatch Logs, CloudWatch Metrics, Bedrock model invocation logging, X-Ray, CloudTrail) and explain the telemetry signals each captures for FM-powered applications.
Implement monitoring dashboards and alerting rules capturing inference latency (P50/P95/P99), token throughput, error rates, guardrail block rates, throttling spikes, and cost overrun thresholds using CloudWatch custom metrics and Bedrock invocation logs.
Define observability strategy that integrates model-level, application-level, and infrastructure-level telemetry into unified dashboards supporting alerting, incident triage, and continuous optimization loops.

Evaluation systems for generative AI quality

Implement automated evaluation pipelines using Bedrock model evaluation jobs, LLM-as-judge patterns, RAGAS metrics, golden dataset comparisons, and human feedback collection for continuous quality measurement across relevance, faithfulness, coherence, and harmfulness dimensions.
Analyze evaluation trends to prioritize prompt refinement, retrieval tuning, and model selection adjustments for sustained production quality based on quality dimension regression patterns.

Troubleshooting generative AI applications

Troubleshoot common Bedrock failures including model access errors, throttling (ThrottlingException), validation errors, context length exceeded, timeout errors, guardrail intervention responses, and agent action group Lambda errors using error codes and CloudWatch logs.
Troubleshoot RAG pipeline failures including knowledge base sync errors, embedding dimension mismatches, retrieval returning irrelevant results, and missing source attribution by analyzing pipeline stage logs and vector store query diagnostics.
Analyze incident patterns across model, retrieval, agent, and integration layers to develop troubleshooting runbooks, post-incident remediation plans, and reliability improvement strategies for generative AI systems.

Certification Benefits

Salary Impact

$172,000

Average Salary

Related Job Roles

GenAI Developer ML Engineer AI Solutions Architect Applied AI Engineer

Industry Recognition

The AWS Generative AI Developer Professional is one of the newest and most forward-looking AWS certifications, targeting the explosive demand for production GenAI skills. As enterprises race to deploy foundation model-powered applications, certified GenAI developers are among the most sought-after technical professionals in the market.

Scope

Included Topics

All domains and task statements in the AWS Certified Generative AI Developer - Professional (AIP-C01) exam guide: Domain 1 Solution Design for Business Applications (30%), Domain 2 Development of Generative AI Applications (26%), Domain 3 Security, Compliance, and Governance (22%), Domain 4 Performance Optimization (12%), and Domain 5 Monitoring and Troubleshooting (10%).
Professional-level design and implementation of production generative AI applications on AWS, including foundation model selection via Amazon Bedrock, retrieval-augmented generation with Knowledge Bases, agentic workflows with Bedrock Agents, prompt engineering, fine-tuning, RLHF, and operational controls.
Service-specific expertise across Amazon Bedrock, SageMaker, Lambda, S3, OpenSearch Serverless, Amazon Kendra, Amazon Lex, Amazon Titan models, Anthropic Claude on Bedrock, Bedrock Guardrails, vector databases, embedding models, LangChain integration, and RAG architecture patterns.
Scenario-based architectural and engineering decisions balancing quality, safety, governance, latency, throughput, resilience, and cost for enterprise generative AI systems.

Not Covered

General cloud architecture coverage not directly tied to generative AI application outcomes in AIP-C01.
Research-only machine learning theory and model pretraining internals that exceed practical FM integration and operations expectations for the certification.
Provider-agnostic implementation details that do not map to AWS services, APIs, and governance mechanisms used in exam objectives.
Short-lived pricing figures, promotional programs, and unstable commercial details that are not durable domain knowledge.

Official Exam Page

Learn more at Amazon Web Services

Visit