# AWS AI Certifications — Course Map

> Full certification roadmap with exam structures, domain weightings, and key topics.
> Last updated: 2026-03-07

---

## Certification Track at a Glance

| # | Certification | Code | Level | Questions | Duration | Pass Score | Cost (USD) | Status |
|---|--------------|------|-------|-----------|----------|------------|------------|--------|
| 1 | AWS Certified AI Practitioner | AIF-C01 | Foundational | 65 (50 scored) | 90 min | 700 / 1000 | $100 | ✅ Active |
| 2 | AWS Certified ML Engineer – Associate | MLA-C01 | Associate | 65 (50 scored) | 130 min | 720 / 1000 | $150 | ✅ Active |
| 3 | AWS Certified GenAI Developer – Professional | AIP-C01 | Professional | 85 (75 scored) | 205 min | ~750 / 1000 | $150 | 🆕 Beta |
| 4 | AWS Certified ML – Specialty | MLS-C01 | Specialty | 65 | 180 min | 750 / 1000 | $300 | ⚠️ Retiring 31 Mar 2026 |

---

## Certification 1 — AWS Certified AI Practitioner (AIF-C01)

**Level:** Foundational | **Validity:** 3 years | **Recommended Experience:** 6 months exposure to AI/ML on AWS

### Domain Breakdown

| Domain   | Topic                                                 | Weight |
| -------- | ----------------------------------------------------- | ------ |
| Domain 1 | Fundamentals of AI and ML                             | 20%    |
| Domain 2 | Fundamentals of Generative AI                         | 24%    |
| Domain 3 | Applications of Foundation Models                     | 28%    |
| Domain 4 | Guidelines for Responsible AI                         | 14%    |
| Domain 5 | Security, Compliance, and Governance for AI Solutions | 14%    |

### Domain 1 — Fundamentals of AI and ML (20%)

- Basic AI/ML concepts and terminology (supervised, unsupervised, reinforcement learning)
- Practical use cases for AI on AWS
- ML development lifecycle (data collection → training → evaluation → deployment)
- Types of ML problems: classification, regression, clustering, forecasting
- Key AWS services: Amazon SageMaker, Amazon Rekognition, Amazon Comprehend, Amazon Forecast

### Domain 2 — Fundamentals of Generative AI (24%)

- What generative AI is and how it differs from traditional ML
- Large Language Models (LLMs): how they work, capabilities and limitations
- Foundation models: GPT-style transformers, diffusion models, multimodal models
- AWS GenAI infrastructure: Amazon Bedrock, Amazon Q, SageMaker JumpStart
- Token limits, context windows, hallucination, model grounding
- Common use cases: text generation, summarisation, code generation, image synthesis

### Domain 3 — Applications of Foundation Models (28%)

- Prompt engineering techniques: zero-shot, few-shot, chain-of-thought
- Retrieval-Augmented Generation (RAG): how it works and when to use it
- Fine-tuning vs. RAG vs. in-context learning — trade-offs
- Foundation model selection criteria: cost, latency, accuracy, modality
- Amazon Bedrock: model selection, knowledge bases, agents
- Evaluating foundation model performance: ROUGE, BLEU, human evaluation
- Design patterns for GenAI applications

### Domain 4 — Guidelines for Responsible AI (14%)

- Responsible AI principles: fairness, explainability, privacy, robustness
- Bias in ML: data bias, algorithmic bias, how to detect and mitigate
- Model explainability tools: Amazon SageMaker Clarify
- AI ethics frameworks and AWS Responsible AI practices
- Transparency, accountability, and human oversight

### Domain 5 — Security, Compliance, and Governance for AI (14%)

- Data security for AI workloads: encryption, access control, IAM
- AWS security services relevant to AI: AWS Macie, AWS KMS, AWS IAM
- Compliance frameworks: GDPR, HIPAA considerations for AI systems
- Model governance: versioning, auditability, monitoring
- Amazon Bedrock guardrails: content filtering, PII redaction

### Key AWS Services — AIF-C01

```
Core AI Services:
├── Amazon SageMaker          — end-to-end ML platform
├── Amazon Bedrock            — foundation models as a service
├── Amazon Q                  — GenAI assistant for business / developers
├── Amazon Rekognition         — image and video analysis
├── Amazon Comprehend          — NLP, sentiment, entity extraction
├── Amazon Polly               — text-to-speech
├── Amazon Transcribe          — speech-to-text
├── Amazon Translate           — language translation
├── Amazon Lex                 — conversational AI / chatbots
├── Amazon Forecast            — time series forecasting
├── Amazon Personalize         — recommendation systems
└── Amazon Textract            — document text and data extraction

Supporting Services:
├── AWS IAM                    — access control
├── Amazon S3                  — data storage
├── Amazon CloudWatch          — monitoring
└── AWS Macie                  — data security and privacy
```

---

## Certification 2 — AWS Certified ML Engineer – Associate (MLA-C01)

**Level:** Associate | **Validity:** 3 years | **Recommended Experience:** 1+ year in ML engineering role on AWS

### Domain Breakdown

| Domain | Topic | Weight |
|--------|-------|--------|
| Domain 1 | Data Preparation for Machine Learning | 28% |
| Domain 2 | ML Model Development | 26% |
| Domain 3 | Deployment and Orchestration of ML Workflows | 22% |
| Domain 4 | ML Solution Monitoring, Maintenance, and Security | 24% |

### Domain 1 — Data Preparation for Machine Learning (28%)

- **Data ingestion:** AWS Glue, Amazon S3, Amazon Kinesis, AWS Data Pipeline
- **Data transformation:** ETL pipelines, feature engineering, normalisation, encoding
- **Feature stores:** Amazon SageMaker Feature Store — online vs. offline stores
- **Data validation:** detecting anomalies, handling missing values, class imbalance
- **Data labelling:** Amazon SageMaker Ground Truth, active learning, labelling strategies
- **Data storage patterns:** data lake (S3), data warehouse (Redshift), streaming (Kinesis)
- **Data governance:** AWS Glue Data Catalog, AWS Lake Formation, lineage tracking

### Domain 2 — ML Model Development (26%)

- **Algorithm selection:** choosing between built-in SageMaker algorithms, custom code, and pre-trained models
- **SageMaker training jobs:** instance types, distributed training, Spot Instance training
- **Hyperparameter tuning:** Amazon SageMaker Automatic Model Tuning (HPO)
- **Model evaluation:** cross-validation, confusion matrix, AUC-ROC, precision/recall
- **Bias and explainability:** SageMaker Clarify for bias detection and SHAP values
- **Experiment tracking:** SageMaker Experiments — comparing runs, metrics, artefacts
- **Model registry:** SageMaker Model Registry — versioning, approval workflows

### Domain 3 — Deployment and Orchestration of ML Workflows (22%)

- **Endpoint types:**
  - Real-time endpoints — low latency, synchronous inference
  - Serverless endpoints — cost-efficient, variable traffic
  - Async endpoints — large payloads, non-urgent workloads
  - Batch Transform — bulk inference, no persistent endpoint
- **Multi-model endpoints:** hosting multiple models on a single endpoint
- **Auto scaling:** target tracking, step scaling for SageMaker endpoints
- **CI/CD for ML:** SageMaker Pipelines, AWS CodePipeline, MLflow integration
- **SageMaker Pipelines:** building reproducible ML workflows (DAG-based)
- **Container management:** SageMaker built-in containers, custom Docker containers
- **Infrastructure as Code:** AWS CloudFormation, CDK for ML infrastructure

### Domain 4 — ML Solution Monitoring, Maintenance, and Security (24%)

- **Model monitoring:** SageMaker Model Monitor — data drift, model quality, bias drift
- **Drift detection:** data drift vs. concept drift — detection strategies
- **Retraining triggers:** scheduled retraining, drift-based retraining pipelines
- **A/B testing:** SageMaker endpoint traffic splitting for model comparison
- **Security best practices:** VPC for SageMaker, encryption at rest and in transit
- **IAM for ML:** least privilege, SageMaker execution roles, resource-based policies
- **Cost optimisation:** right-sizing instances, Spot Training, Savings Plans

### Key AWS Services — MLA-C01

```
Data Layer:
├── Amazon S3                      — data lake storage
├── AWS Glue                       — ETL and data catalogue
├── Amazon Kinesis                 — real-time data streaming
├── AWS Lake Formation             — data lake governance
└── Amazon SageMaker Feature Store — centralised feature management

Model Development:
├── Amazon SageMaker Studio        — unified ML IDE
├── SageMaker Training Jobs        — managed training infrastructure
├── SageMaker Automatic Model Tuning — HPO at scale
├── SageMaker Clarify              — bias detection and explainability
└── SageMaker Experiments          — experiment tracking

Deployment:
├── SageMaker Real-time Endpoints  — synchronous inference
├── SageMaker Batch Transform      — bulk inference
├── SageMaker Serverless Inference — event-driven inference
└── SageMaker Pipelines            — ML workflow orchestration

Operations:
├── SageMaker Model Monitor        — production monitoring
├── SageMaker Model Registry       — model versioning and governance
├── Amazon CloudWatch              — logging and alerting
└── AWS CloudTrail                 — audit logging
```

---

## Certification 3 — AWS Certified GenAI Developer – Professional (AIP-C01)

**Level:** Professional | **Validity:** 3 years | **Recommended Experience:** 2+ years building production apps, 1+ year implementing GenAI solutions | **Status:** Beta (available through March 2026 beta period)

### Domain Breakdown

| Domain | Topic | Weight |
|--------|-------|--------|
| Domain 1 | Foundation Model Integration, Data Management, and Compliance | 31% |
| Domain 2 | Implementation and Integration | 26% |
| Domain 3 | AI Safety, Security, and Governance | 20% |
| Domain 4 | Operational Efficiency and Optimisation for GenAI Applications | 12% |
| Domain 5 | Testing, Validation, and Troubleshooting | 11% |

### Domain 1 — Foundation Model Integration, Data Management, and Compliance (31%)

- **Foundation model selection:** evaluating models in Amazon Bedrock by task, cost, performance
- **RAG architecture:** design patterns, chunking strategies, embedding models
- **Vector stores:** Amazon OpenSearch Service, Amazon Aurora (pgvector), Pinecone integration
- **Knowledge bases:** Amazon Bedrock Knowledge Bases — ingestion, retrieval, grounding
- **Data preparation for GenAI:** cleaning, chunking, metadata tagging for retrieval
- **Compliance considerations:** data residency, PII handling, regulatory requirements
- **Model customisation:** fine-tuning on Bedrock, continued pre-training

### Domain 2 — Implementation and Integration (26%)

- **Agentic AI:** Amazon Bedrock Agents — tool use, action groups, function calling
- **Prompt engineering advanced:** system prompts, prompt templates, dynamic prompting
- **LangChain and similar frameworks:** integration patterns with Amazon Bedrock
- **Multi-step reasoning:** chain-of-thought, ReAct patterns, tool-use agents
- **Application integration:** API Gateway + Lambda + Bedrock patterns
- **Streaming responses:** invokeModelWithResponseStream API
- **Context management:** conversation history, memory patterns, context window limits

### Domain 3 — AI Safety, Security, and Governance (20%)

- **Amazon Bedrock Guardrails:** content filtering, denied topics, PII redaction, grounding checks
- **Prompt injection attacks:** detection and mitigation strategies
- **Model access control:** IAM policies for Bedrock, cross-account access
- **Output safety:** toxicity detection, factual grounding, hallucination mitigation
- **Audit and logging:** AWS CloudTrail for Bedrock, model invocation logging
- **Responsible AI at scale:** governance frameworks, model cards, impact assessments

### Domain 4 — Operational Efficiency and Optimisation (12%)

- **Cost optimisation:** provisioned throughput vs. on-demand in Bedrock
- **Latency optimisation:** caching strategies, streaming, parallel processing
- **Provisioned throughput:** when to use, pricing model, commitment terms
- **Batch inference:** Amazon Bedrock batch API for high-volume workloads
- **Prompt caching:** reducing token costs for repeated context

### Domain 5 — Testing, Validation, and Troubleshooting (11%)

- **Evaluation frameworks:** RAGAS, LLM-as-a-judge, human evaluation
- **A/B testing GenAI:** comparing model versions, prompt variants
- **Debugging GenAI:** tracing, logging, Bedrock model invocation logs
- **Performance testing:** throughput, latency benchmarking
- **Regression testing:** validating model behaviour after updates

### Key AWS Services — AIP-C01

```
Foundation Model Platform:
├── Amazon Bedrock                 — managed foundation models
├── Amazon Bedrock Knowledge Bases — RAG with managed vector stores
├── Amazon Bedrock Agents          — agentic AI with tool use
├── Amazon Bedrock Guardrails      — safety and content filtering
└── Amazon Bedrock Flows           — visual GenAI workflow builder

Vector and Search:
├── Amazon OpenSearch Serverless   — vector search
├── Amazon Aurora (pgvector)       — relational vector store
└── Amazon MemoryDB                — in-memory vector caching

Integration:
├── AWS Lambda                     — serverless compute for GenAI apps
├── Amazon API Gateway             — REST/WebSocket API for GenAI
├── Amazon EventBridge             — event-driven GenAI workflows
└── AWS Step Functions             — complex workflow orchestration

Observability:
├── Amazon CloudWatch              — metrics, logs, alarms
├── AWS CloudTrail                 — audit trail for Bedrock
└── AWS X-Ray                      — distributed tracing
```

---

## Retired / Retiring Certifications

### AWS Certified ML – Specialty (MLS-C01) ⚠️

- **Last exam date:** 31 March 2026
- **Replacement:** MLA-C01 (Associate) for engineering focus; AIP-C01 (Professional) for advanced GenAI
- **Action:** Do **not** pursue this certification — pivot to MLA-C01 as the current path

---

## Exam Logistics — All Certifications

| Aspect | Details |
|--------|---------|
| Testing options | Pearson VUE testing centre or online proctored |
| Retake policy | 14-day wait after first failed attempt |
| Validity | 3 years from passing date |
| Recertification | Retake current exam or pass higher-level exam |
| Accommodations | Extra time available — request via Pearson VUE |
| Score report | Available immediately after online exam; scaled 100–1000 |
| Digital badge | Credly badge issued upon passing |

---

*Tags: #AWS #AI #MachineLearning #AIF-C01 #MLA-C01 #AIP-C01 #Certifications #CourseMap*