How vicsz/aif-c01-study-notes works
1. This repository is not an application or service codebase. It is a small documentation repository centered on AWS AI/ML exam preparation content for the AWS Certified AI Practitioner (AIF-C01).
2. The repository has two primary artifacts:
- README.md: the main human-readable source of truth and navigation hub.
- aws-aif-c01-study-sheet.pdf: a compiled study sheet artifact, likely exported for offline reading or distribution.
3. The main boundary is between editable markdown content and a packaged PDF deliverable. The README acts as the authoritative, structured knowledge base, while the PDF is a consumable snapshot of that material.
4. The content is organized around conceptual learning domains rather than software components:
- AI/ML fundamentals
- Generative AI and LLM concepts
- Prompting and embeddings
- ML lifecycle and MLOps
- Governance, security, and cost optimization
- AWS managed AI services
- AWS platform services used by ML workflows
5. A major subsystem in the notes is the ML pipeline. It is described as a flow from business goal definition through problem framing, data collection, preprocessing, feature engineering, training/tuning, deployment, and monitoring. This is the closest thing to an end-to-end architecture in the repo.
6. Data flows in the study material are mostly conceptual:
- Source systems, APIs, and third-party datasets feed data into Amazon S3.
- AWS Glue, DataBrew, Macie, and SageMaker tools transform and prepare that data.
- Training jobs consume curated data from S3.
- Model artifacts are written back to S3 after training.
- Deployed models are served through SageMaker endpoints or Bedrock inference paths.
7. The repo distinguishes two broad AWS AI execution paths:
- Amazon SageMaker for custom ML lifecycle management, training, deployment, monitoring, and governance.
- Amazon Bedrock for managed foundation-model and generative-AI usage, including RAG, agents, guardrails, and provisioned throughput for custom models.
8. Another important boundary is between high-level managed AI services and custom ML platforms:
- Managed services like Rekognition, Comprehend, Textract, Lex, Transcribe, Polly, Forecast, Personalize, and Translate are positioned as task-specific APIs.
- SageMaker is positioned as the flexible platform for building and operating custom models.
- Bedrock is positioned as the managed foundation-model layer for GenAI applications.
9. Retrieval and vector search form a separate conceptual subsystem. The notes treat embeddings, semantic search, and vector databases as the retrieval layer for RAG-style systems, with AWS options including OpenSearch, Aurora/RDS with pgvector, MemoryDB, DocumentDB, and Neptune-related capabilities.
10. Governance, explainability, and compliance are treated as first-class architectural concerns. The repository groups SageMaker Clarify, Model Cards, lineage tracking, Audit Manager, Artifact, CloudTrail, Config, and security controls like IAM, KMS, PrivateLink, and S3 Block Public Access as the trust-and-control layer around ML systems.
11. Monitoring and retraining are another lifecycle boundary. The notes separate deployment from ongoing operation by using SageMaker Model Monitor, CloudWatch integration, drift detection, and retraining pipelines to represent the feedback loop after a model is live.
12. The repository also includes a cost/performance optimization layer. Managed Spot Training, instance selection, Profiler, Trusted Advisor, and inference options are documented as operational levers that affect workload efficiency without changing the core ML logic.
13. From an information-architecture perspective, the README is structured as a layered study guide: definitions first, then pipeline stages, then service catalogs, then comparison tables and cheat sheets. The PDF likely mirrors or packages this same layered content for offline use.
14. If you are drawing an architecture diagram for this repository, the most accurate depiction is a documentation system with one source document feeding a compiled reference artifact, not a runtime system with live components. The “architecture” lives in the conceptual relationships among AWS AI services, the ML lifecycle, and governance/operational controls described in the notes.
Main entry points: README.md README · aws-aif-c01-study-sheet.pdf Study sheet
Components 16
Repository docs
Primary authored and packaged study materials.
- README · Main source of truth and navigation entry for the study notes.
README.md - Study sheet · Packaged PDF snapshot of the study guide for offline use.
aws-aif-c01-study-sheet.pdf
ML lifecycle
End-to-end machine learning workflow and operating loop.
- AI basics · Core AI and ML definitions, terminology, and foundational concepts.
- ML pipeline · Business goal, data prep, training, deployment, and monitoring flow.
- Data prep · Collection, cleaning, preprocessing, and feature engineering.
- Training & tuning · Model training, hyperparameter tuning, and artifact creation.
- Deployment · Serving models for inference through managed endpoints or GenAI paths.
AWS AI services
Managed AI, GenAI, and platform services referenced in the notes.
- GenAI & LLMs · Generative AI, foundation models, and large language model concepts.
- Prompts & embeddings · Prompting, embeddings, semantic similarity, and RAG prerequisites.
- SageMaker · Flexible platform for custom model building, deployment, and operations.
- Bedrock · Managed foundation-model platform for GenAI apps, agents, and RAG.
- AI APIs · Specialized AWS AI services such as Rekognition, Comprehend, Textract, Lex, Transcribe, Polly, Forecast, Personalize, and Translate.
- Vector search · Embeddings, semantic search, and vector stores used for RAG.
Control layer
Security, governance, monitoring, and optimization concerns.
- Monitoring loop · Drift detection, metrics, and retraining feedback after release.
- Governance · Explainability, lineage, audit, security, and compliance controls.
- Optimization · Cost and performance levers such as spot training, profiling, and instance choices.
Connections 22
- README → Study sheet: packages · The README is the editable source, and the PDF is the distributable snapshot.
- README → AI basics: introduces · The notes begin with foundational AI and ML concepts.
- README → ML pipeline: organizes · The guide is structured around the ML lifecycle.
- README → SageMaker: covers · The notes explain SageMaker as the custom ML platform.
- README → Bedrock: covers · The notes explain Bedrock as the managed GenAI platform.
- README → AI APIs: catalogs · The guide catalogs task-specific AWS AI services.
- README → Vector search: explains · The guide includes embeddings, vector search, and RAG patterns.
- README → Governance: includes · Governance, compliance, and security are treated as first-class topics.
- README → Optimization: includes · The guide includes operational cost and performance tuning.
- ML pipeline → Data prep: starts with · Pipeline work begins by collecting and preparing data.
- Data prep → Training & tuning: feeds · Curated data is passed into training and tuning jobs.
- Training & tuning → Deployment: produces · Training yields model artifacts that are deployed for inference.
- Deployment → Monitoring loop: loops to · Production use feeds monitoring, drift detection, and retraining.
- SageMaker → ML pipeline: implements · SageMaker supports the full custom ML lifecycle.
- Bedrock → GenAI & LLMs: implements · Bedrock provides the managed foundation-model runtime.
- Bedrock → Vector search: uses · RAG on Bedrock depends on embeddings and vector retrieval.
- Vector search → Prompts & embeddings: depends on · Retrieval is driven by embeddings and semantic similarity.
- AI APIs → AI basics: applies to · Task-specific services are framed as applied AI capabilities.
- Governance → SageMaker: wraps · Security, lineage, and explainability controls surround SageMaker workflows.
- Governance → Bedrock: wraps · Governance and security controls also apply to GenAI usage.
- Optimization → Training & tuning: tunes · Spot training, profiling, and instance selection improve training efficiency.
- Optimization → Deployment: tunes · Inference options and sizing affect serving cost and latency.