# AWS AI Certifications — Progressive Study Plan

> **Starting Level:** Beginner (no prior AWS AI/ML experience)
> **Target:** AIF-C01 → MLA-C01 → AIP-C01
> **Total Duration:** ~7 months (can compress to 5 months with full-time study)
> **Daily Commitment:** 1–2 hours on weekdays, 3–4 hours on weekends

---

## Overview — Three-Phase Plan

```
PHASE 1           PHASE 2              PHASE 3
─────────         ─────────────        ─────────────────────
Weeks 1–8         Weeks 9–20           Weeks 21–32
AIF-C01           MLA-C01              AIP-C01
AI Practitioner   ML Engineer Assoc.   GenAI Developer Pro
~2 months         ~3 months            ~3 months
```

---

## Phase 1 — AIF-C01: AWS Certified AI Practitioner

**Target exam date:** ~8 weeks from start date
**Estimated study hours:** 60–80 hours total

### Week 1 — AI/ML Foundations

**Goal:** Build conceptual foundation in AI/ML before touching AWS tools.

| Day | Topic | Resource | Time |
|-----|-------|---------|------|
| Mon | AI vs ML vs Deep Learning — core distinctions | [[AWS/AWS-AI-Certs/00-Overview]] Domain 1 | 1 hr |
| Tue | Supervised learning — classification and regression | AWS Skill Builder: ML Fundamentals | 1 hr |
| Wed | Unsupervised learning — clustering and anomaly detection | AWS Skill Builder: ML Fundamentals | 1 hr |
| Thu | ML lifecycle — data → training → evaluation → deployment | [[AWS/AWS-AI-Certs/00-Overview]] Domain 1 | 1 hr |
| Fri | Evaluation metrics — accuracy, precision, recall, F1, AUC-ROC | [[Cheatsheets/ML-Fundamentals-Reference]] | 1 hr |
| Sat | Hands-on: AWS Free Tier — create an AWS account, explore console | AWS Console | 2 hrs |
| Sun | Review Week 1, update notes, quiz yourself | Self-test | 1 hr |

**Week 1 Checklist:**
- [x] Understand the difference between AI, ML, deep learning, and generative AI
- [x] Know the three main ML paradigms and at least two examples of each
- [x] Understand the complete ML lifecycle
- [ ] Can explain precision vs. recall trade-off in plain English
- [x] AWS account created and console familiarised

---

### Week 2 — Generative AI and Foundation Models

**Goal:** Understand how GenAI and LLMs work conceptually.

| Day | Topic | Resource | Time |
|-----|-------|---------|------|
| Mon | What is a foundation model — architecture, training at scale | [[AWS/AWS-AI-Certs/00-Overview]] Domain 2 | 1 hr |
| Tue | How LLMs generate text — tokenisation, transformer, next-token prediction | [[AWS/AWS-AI-Certs/00-Overview]] Domain 2 | 1 hr |
| Wed | Tokens, context windows, temperature, top-P — parameter effects | [[Cheatsheets/GenAI-Reference]] | 1 hr |
| Thu | Hallucination — what causes it, how to mitigate | [[AWS/AWS-AI-Certs/00-Overview]] Domain 2 | 1 hr |
| Fri | Amazon Bedrock overview — model providers, console exploration | AWS Skill Builder: Bedrock Intro | 1 hr |
| Sat | Hands-on: Amazon Bedrock playground — try Claude, Titan, Llama | AWS Console (Bedrock) | 3 hrs |
| Sun | Review + self-test on GenAI concepts | Flashcards | 1 hr |

**Week 2 Checklist:**
- [ ] Can explain how an LLM generates output step by step
- [ ] Understand the trade-off between temperature and output quality
- [ ] Know what hallucination is and three mitigation strategies
- [ ] Have explored Amazon Bedrock in the console

---

### Week 3 — Prompt Engineering and RAG

**Goal:** Master prompt techniques and understand RAG architecture thoroughly (Domain 3 = 28%).

| Day | Topic | Resource | Time |
|-----|-------|---------|------|
| Mon | Prompt engineering — zero-shot, few-shot, chain-of-thought | [[AWS/AWS-AI-Certs/00-Overview]] Domain 3 | 1 hr |
| Tue | System prompts, role prompting, output formatting | [[Cheatsheets/GenAI-Reference]] | 1 hr |
| Wed | RAG — the problem it solves, architecture overview | [[AWS/AWS-AI-Certs/00-Overview]] Domain 3 | 1 hr |
| Thu | RAG deep dive — chunking, embedding, vector search, retrieval | [[AWS/AWS-AI-Certs/00-Overview]] Domain 3 | 1 hr |
| Fri | RAG vs. Fine-tuning vs. in-context learning — when to use what | [[AWS/AWS-AI-Certs/00-Overview]] Domain 3 | 1 hr |
| Sat | Hands-on: Build a simple RAG in Bedrock Knowledge Bases | AWS Console (Bedrock) | 3 hrs |
| Sun | Foundation model evaluation — BLEU, ROUGE, BERTScore | [[AWS/AWS-AI-Certs/Course-Map]] Domain 3 | 1 hr |

**Week 3 Checklist:**
- [ ] Can describe zero-shot, few-shot, and chain-of-thought with examples
- [ ] Can draw the full RAG architecture from memory (indexing + retrieval + generation)
- [ ] Know when to choose RAG vs. fine-tuning vs. prompting alone
- [ ] Have created a Knowledge Base in Amazon Bedrock

---

### Week 4 — AWS AI Services

**Goal:** Know all the purpose-built AWS AI services and their use cases.

| Day | Topic | Resource | Time |
|-----|-------|---------|------|
| Mon | Amazon Rekognition, Comprehend, Polly, Transcribe, Translate | [[Cheatsheets/AWS-AI-Services-Reference]] | 1 hr |
| Tue | Amazon Lex, Kendra, Textract, Forecast, Personalize | [[Cheatsheets/AWS-AI-Services-Reference]] | 1 hr |
| Wed | Amazon SageMaker — overview, Studio, what it can do | [[AWS/AWS-AI-Certs/Course-Map]] MLA section | 1 hr |
| Thu | Amazon Q — Q Business vs Q Developer, use cases | AWS Documentation | 1 hr |
| Fri | Service selection scenarios — matching use case to service | [[Cheatsheets/Exam-Tips-Strategies]] | 1 hr |
| Sat | Hands-on: Explore 3 AI services in AWS Console | AWS Console | 3 hrs |
| Sun | Review + service matching quiz | Self-test | 1 hr |

**Week 4 Checklist:**
- [ ] Can match any business scenario to the correct AWS AI service
- [ ] Know the difference between Amazon Q Business and Q Developer
- [ ] Understand SageMaker's role vs. purpose-built AI services
- [ ] Have used at least 3 AWS AI services hands-on

---

### Week 5 — Responsible AI and Governance

**Goal:** Cover Domains 4 and 5 — responsible AI, security, compliance.

| Day | Topic | Resource | Time |
|-----|-------|---------|------|
| Mon | Responsible AI pillars — fairness, explainability, privacy, robustness | [[AWS/AWS-AI-Certs/00-Overview]] Domain 4 | 1 hr |
| Tue | Bias types — data bias, algorithmic bias, measurement bias | [[AWS/AWS-AI-Certs/00-Overview]] Domain 4 | 1 hr |
| Wed | SageMaker Clarify — bias detection, SHAP explanations | [[AWS/AWS-AI-Certs/Course-Map]] Domain 4 | 1 hr |
| Thu | AI security — IAM, KMS, VPC, Macie for AI workloads | [[AWS/AWS-AI-Certs/00-Overview]] Domain 5 | 1 hr |
| Fri | Bedrock Guardrails — content filtering, PII, denied topics | [[AWS/AWS-AI-Certs/00-Overview]] Domain 5 | 1 hr |
| Sat | Compliance — GDPR, HIPAA considerations, model governance | [[AWS/AWS-AI-Certs/Course-Map]] Domain 5 | 2 hrs |
| Sun | Review Domains 4 and 5, quiz | Self-test | 1 hr |

**Week 5 Checklist:**
- [ ] Know all six responsible AI pillars and the AWS tool for each
- [ ] Can describe at least four types of AI/ML bias
- [ ] Understand all five Bedrock Guardrail policy types
- [ ] Can design a secure AI architecture using IAM, KMS, and VPC

---

### Week 6 — Full Domain Review

**Goal:** Review all five domains, identify weak areas.

| Day | Activity |
|-----|---------|
| Mon | Re-read [[AWS/AWS-AI-Certs/00-Overview]] Domain 1 and 2, make flashcards |
| Tue | Re-read [[AWS/AWS-AI-Certs/00-Overview]] Domain 3, draw RAG diagram from memory |
| Wed | Re-read Domains 4 and 5, review security service table |
| Thu | Take first practice exam (Tutorials Dojo or AWS Official) |
| Fri | Review all incorrect answers — understand why each is wrong |
| Sat | Deep dive on weakest domain from practice exam results |
| Sun | Second practice exam (different question set) |

**Week 6 Goal:** Score 75%+ on practice exams before attempting real exam.

---

### Weeks 7–8 — Final Prep and Exam

| Week | Activity |
|------|---------|
| Week 7 | 3× practice exams (fresh sets), review wrong answers daily |
| Week 7 | Study [[Cheatsheets/Exam-Tips-Strategies]] — exam technique |
| Week 7 | Review [[Exam-Prep/AIF-C01-Checklist]] |
| Week 8 Day 1-3 | Light review — flashcards and weak areas only |
| Week 8 Day 4 | Rest day — no heavy studying |
| Week 8 Day 5 | ✅ **AIF-C01 EXAM** |

**Practice Exam Target:** Score 80%+ consistently before booking exam.

---

## Phase 2 — MLA-C01: AWS Certified ML Engineer – Associate

**Prerequisite:** AIF-C01 passed ✅
**Target exam date:** ~12 weeks after AIF-C01
**Estimated study hours:** 100–120 hours total
**Use 50% voucher from AIF-C01**

### Week 1 (of Phase 2) — Data Engineering for ML

**Goal:** Master the data layer — the highest-weighted domain (28%).

| Topic | Deep Dive Items |
|-------|----------------|
| AWS Glue | Crawlers, jobs, data catalogue, dynamic frames, partitioning |
| Amazon S3 | Storage classes for ML data, lifecycle policies, S3 Select |
| Amazon Kinesis | Data Streams vs Firehose vs Analytics — when to use each |
| AWS Lake Formation | Column-level permissions, cross-account data sharing |
| SageMaker Feature Store | Online vs offline feature groups, ingestion, retrieval |

**Hands-on:** Build a full data pipeline: Kinesis → Lambda → S3 → Glue → SageMaker Feature Store.

---

### Weeks 2–3 (of Phase 2) — SageMaker Training and Model Development

**Goal:** Deep SageMaker expertise — training, HPO, evaluation.

| Topic | Key Items to Master |
|-------|-------------------|
| Training jobs | Container options, instance selection, distributed training |
| Spot training | checkpointing, max_wait, cost savings calculation |
| SageMaker built-in algorithms | XGBoost, Linear Learner, K-Means, BlazingText, DeepAR |
| SageMaker Experiments | Run tracking, comparison, artefact management |
| SageMaker Debugger | Built-in rules, profiling, hook configuration |
| HPO (Automatic Model Tuning) | Bayesian vs Random vs Hyperband, objective metrics |
| SageMaker Clarify (training) | Bias report, SHAP explainability on training data |

**Hands-on:** Train an XGBoost model on SageMaker with HPO. Compare 10 runs in Experiments.

---

### Weeks 4–5 (of Phase 2) — Deployment and MLOps

**Goal:** Master SageMaker deployment options and ML pipeline automation.

| Topic | Key Items to Master |
|-------|-------------------|
| Endpoint types | Real-time, Serverless, Async, Batch Transform — decision matrix |
| Multi-model endpoints | Use cases, cold start, eviction, invocation |
| Auto scaling | Target tracking policies for SageMaker endpoints |
| SageMaker Pipelines | Steps types, conditions, parameter overrides |
| Model Registry | Approval workflows, model packages, cross-account |
| CI/CD for ML | CodePipeline + SageMaker Pipelines integration pattern |
| Containers | BYO containers, inference containers, multi-container |

**Hands-on:** Build a SageMaker Pipeline with training → evaluation → conditional registration → deployment.

---

### Weeks 6–7 (of Phase 2) — Monitoring, Security, and Cost

**Goal:** Cover Domain 4 (24%) — ML operations and security.

| Topic | Key Items to Master |
|-------|-------------------|
| SageMaker Model Monitor | Data quality, model quality, bias drift, feature attribution drift |
| Drift detection | PSI, KS test, threshold configuration |
| Retraining pipelines | EventBridge trigger → Pipeline execution → re-deploy |
| SageMaker security | VPC mode, encryption, inter-container traffic encryption |
| IAM for ML | Execution roles, S3 bucket policies, cross-account patterns |
| Cost optimisation | Right-sizing, Spot instances, Savings Plans, endpoint scaling |

**Hands-on:** Set up SageMaker Model Monitor on a deployed endpoint with CloudWatch alerts.

---

### Weeks 8–10 (of Phase 2) — Full Review and Practice Exams

| Week | Activity |
|------|---------|
| Week 8 | Full review of all four domains |
| Week 9 | 3× practice exams — review incorrect answers daily |
| Week 10 | Final prep → [[Exam-Prep/MLA-C01-Checklist]] → **MLA-C01 EXAM** |

**Practice Exam Target:** Score 78%+ consistently before booking.

---

## Phase 3 — AIP-C01: AWS Certified GenAI Developer – Professional

**Prerequisite:** MLA-C01 passed ✅
**Target exam date:** ~12 weeks after MLA-C01
**Estimated study hours:** 100–130 hours total
**Use 50% voucher from MLA-C01**

### Weeks 1–2 (of Phase 3) — Amazon Bedrock Mastery

**Goal:** Deep expertise in Amazon Bedrock — the core of AIP-C01.

| Topic | Key Items to Master |
|-------|-------------------|
| Bedrock APIs | InvokeModel, ConverseAPI, streaming, batch |
| Model selection | Choosing between Claude, Titan, Llama, Mistral, Cohere for task |
| Provisioned throughput | When to buy, model units, commitment pricing |
| Bedrock Knowledge Bases | Chunking strategies, embedding model selection, vector store options |
| Bedrock Agents | Action groups (Lambda/OpenAPI), knowledge bases, session attributes |
| Bedrock Flows | Visual workflow builder for multi-step GenAI workflows |
| Bedrock Guardrails | All five policy types, configuration, monitoring |
| Fine-tuning on Bedrock | Continued pre-training, fine-tuning job setup |

**Hands-on:** Build a Bedrock Agent with a Lambda action group that queries a real API.

---

### Weeks 3–4 (of Phase 3) — Advanced RAG and GenAI Architecture

**Goal:** Production RAG patterns, vector stores, evaluation.

| Topic | Key Items to Master |
|-------|-------------------|
| Advanced chunking | Hierarchical, semantic chunking with Lambda |
| Vector stores | OpenSearch Serverless vs Aurora pgvector vs external — decision criteria |
| Metadata filtering | Filtering vector search results by document metadata |
| Hybrid search | Combining vector search + keyword search (BM25) |
| RAG evaluation | RAGAS metrics — faithfulness, relevance, context precision/recall |
| LLM-as-a-judge | Evaluation framework using strong model to grade weak model |
| Prompt caching | Amazon Bedrock prompt caching — implementation and savings |

---

### Weeks 5–6 (of Phase 3) — AI Safety, Security, and Governance

**Goal:** Master Domain 3 (20%) — enterprise-grade AI safety.

| Topic | Key Items to Master |
|-------|-------------------|
| Prompt injection attacks | Taxonomy, detection strategies, defence patterns |
| Bedrock Guardrails deep dive | All policy types, grounding checks, apply to agents |
| IAM for Bedrock | Model access policies, VPC endpoints, cross-account |
| Audit and compliance | CloudTrail for Bedrock, model invocation logging |
| Responsible AI at scale | Governance frameworks, model cards, impact assessments |
| MITRE ATLAS | AI threat framework — key attack patterns |

---

### Weeks 7–8 (of Phase 3) — Cost Optimisation and Evaluation

**Goal:** Domains 4 and 5 — optimise and test GenAI applications.

| Topic | Key Items to Master |
|-------|-------------------|
| Cost modelling | On-demand vs provisioned throughput — break-even calculation |
| Prompt optimisation | Reducing token count while preserving quality |
| Batch inference | When batch beats on-demand, Bedrock batch API |
| A/B testing GenAI | Traffic splitting, metric collection, statistical significance |
| Debugging GenAI | Trace analysis in Bedrock Agents, invocation logs |
| Performance testing | Throughput limits, concurrency, latency P99 |

---

### Weeks 9–10 (of Phase 3) — Full Review and Final Prep

| Week | Activity |
|------|---------|
| Week 9 | Full domain review, 3× practice exams |
| Week 10 | Final prep → [[Exam-Prep/AIP-C01-Checklist]] → **AIP-C01 EXAM** |

**Practice Exam Target:** Score 78%+ consistently before booking.

---

## Recommended Learning Resources

### Free Resources

| Resource | What It Covers | Link |
|---------|---------------|------|
| AWS Skill Builder (free tier) | Official AWS training courses | skillbuilder.aws |
| AWS Documentation | Service-level deep dives | docs.aws.amazon.com |
| AWS Well-Architected Framework | Architecture best practices | aws.amazon.com/architecture |
| AWS re:Invent Videos (YouTube) | Deep dives by AWS engineers | YouTube: AWS |
| AWS Blogs | Service announcements, how-tos | aws.amazon.com/blogs |

### Paid Resources (Recommended)

| Resource | Best For | Cost |
|---------|---------|------|
| AWS Skill Builder Individual (subscription) | Official practice exams, labs | ~$29/month |
| Tutorials Dojo Practice Exams | Best practice exams for AWS | ~$15–20 per exam set |
| Udemy courses (Stephane Maarek) | Structured video learning | £12–15 on sale |
| A Cloud Guru / Pluralsight | Lab environments | Subscription |

### Hands-On Labs

**For AIF-C01:**
- AWS Skill Builder: "Foundations of Prompt Engineering" (free)
- AWS Skill Builder: "Amazon Bedrock Getting Started" (free)
- Explore Bedrock playground with every model

**For MLA-C01:**
- AWS Skill Builder: "Getting Started with Amazon SageMaker Studio" (free)
- Build an end-to-end ML pipeline from scratch in SageMaker
- Complete the "ML Engineer Learning Plan" on Skill Builder

**For AIP-C01:**
- Build a production RAG application using Bedrock Knowledge Bases
- Create a Bedrock Agent with real API integration
- Implement Guardrails with all five policy types

---

## Weekly Progress Tracker

Use this to monitor your progress through each phase:

### Phase 1 — AIF-C01

| Week | Focus | Done | Score |
|------|-------|------|-------|
| 1 | AI/ML Foundations | ☐ | — |
| 2 | GenAI and Foundation Models | ☐ | — |
| 3 | Prompt Engineering and RAG | ☐ | — |
| 4 | AWS AI Services | ☐ | — |
| 5 | Responsible AI and Governance | ☐ | — |
| 6 | Full Review | ☐ | Practice: —% |
| 7 | Practice Exams | ☐ | Practice: —% |
| 8 | Final Prep + EXAM | ☐ | Actual: —/1000 |

### Phase 2 — MLA-C01

| Week | Focus | Done | Score |
|------|-------|------|-------|
| 1 | Data Engineering | ☐ | — |
| 2–3 | SageMaker Training | ☐ | — |
| 4–5 | Deployment and MLOps | ☐ | — |
| 6–7 | Monitoring and Security | ☐ | — |
| 8–9 | Practice Exams | ☐ | Practice: —% |
| 10 | Final Prep + EXAM | ☐ | Actual: —/1000 |

### Phase 3 — AIP-C01

| Week | Focus | Done | Score |
|------|-------|------|-------|
| 1–2 | Amazon Bedrock Mastery | ☐ | — |
| 3–4 | Advanced RAG | ☐ | — |
| 5–6 | AI Safety and Governance | ☐ | — |
| 7–8 | Cost Optimisation and Evaluation | ☐ | — |
| 9–10 | Practice Exams + EXAM | ☐ | Actual: —/1000 |

---

*Tags: #AWS #AI #StudyPlan #AIF-C01 #MLA-C01 #AIP-C01 #Beginner*