How sunsikim/aws-mla-c01 works
This repository is a documentation-only course notes repo for the AWS Machine Learning Associate certification. There is no application runtime or executable service layer; the “architecture” here is the learning path itself, organized around the end-to-end ML lifecycle on AWS.
1. Top-level structure
The repo is split into four major parts that map directly to the ML workflow:
- data preparation
- model preparation
- model deployment
- pipeline management
This gives it a pipeline-oriented architecture rather than a product/code architecture.
2. Part 1: Data preparation
part1-data-preparation covers the upstream data layer. It is divided into:
- ingestion and storage
- data transformation
This is the foundation for everything else: raw data is brought in, persisted, then prepared for analytics or training.
3. Ingestion boundary
In chapter1-ingestion-and-storage/section1-ingestion-kinesis, the focus is on streaming ingestion through AWS Kinesis. Architecturally, this is the entry point for real-time or near-real-time data feeds before they land in durable storage.
4. Storage boundary
Storage is split across object and file/block services:
- section2-storage-s3 for durable object storage and data lakes
- section3-storage-ebs-efs-fsx for compute-attached or shared file/storage systems
This reflects a key design boundary: S3 for centralized, scalable datasets; EBS/EFS/FSx for workloads needing mounted storage semantics.
5. Transformation layer
part1-data-preparation/chapter2-data-transformation has two concerns:
- querying
- wrangling
This layer consumes data from storage and prepares it for downstream ML use, separating data access/analysis from data cleaning and feature shaping.
6. Part 2: Model preparation
part2-model-preparation moves from data to modeling. It includes:
- AWS managed services
- model training
- generative AI
Architecturally, this is where prepared data becomes training inputs, and where AWS-managed ML capabilities are introduced as abstractions over infrastructure.
7. Training subsystem
chapter2-model-training is split between:
- modeling theories
- SageMaker training
This indicates a mix of conceptual ML topics and AWS-specific training execution. SageMaker is the primary managed training boundary in the repo’s architecture.
8. Generative AI as a separate concern
chapter3-generative-ai is isolated from standard training because it introduces a distinct modeling paradigm and likely different AWS services and patterns. Its separate placement suggests the repo treats GenAI as an extension of the ML lifecycle, not just another training variant.
9. Part 3: Model deployment
part3-model-deployment covers serving models in two ways:
- chapter1-docker-endpoint
- chapter2-sagemaker-endpoint
This shows a deployment boundary between container-based serving and managed SageMaker inference endpoints.
10. Deployment patterns
The Docker endpoint section implies self-managed or containerized inference infrastructure, while the SageMaker endpoint section represents AWS-managed model serving. Together, they frame the tradeoff between operational control and managed simplicity.
11. Part 4: Pipeline management
part4-pipeline-management handles production concerns:
- security
- monitoring
These are cross-cutting controls that sit around ingestion, training, and deployment rather than inside one stage. They define the operational envelope for an ML system on AWS.
12. System flow implied by the repo
The overall flow is:
data ingestion → storage → querying/wrangling → training → deployment → security/monitoring
In AWS terms, the repo emphasizes Kinesis and S3 early, SageMaker for training and serving, and operational governance as the final layer.
13. What this repo is not
This is not a service implementation repo, not a library, and not an application with runtime components. It is structured course material that documents an ML platform architecture on AWS.
14. Diagramming takeaway
If you draw an architecture diagram from this repo, it should center on a data-to-model-to-serving pipeline with AWS services as the major infrastructure blocks, and with security/monitoring as a surrounding operational layer.
Main entry points: part2-model-preparation Model Prep · README.md Course Notes · chapter1-ingestion-and-storage Ingest & Store · part1-data-preparation Data Prep
Components 21
Part 1
Data preparation
- Data Prep · Upstream data preparation for ML workflows
part1-data-preparation - Ingest & Store · Data ingress and durable storage boundaries
part1-data-preparation/chapter1-ingestion-and-storage - Kinesis Ingest · Streaming data entry point
part1-data-preparation/chapter1-ingestion-and-storage/section1-ingestion-kinesis - S3 Storage · Central durable storage and data lake layer
part1-data-preparation/chapter1-ingestion-and-storage/section2-storage-s3 - EBS/EFS/FSx · Mounted or shared storage options
part1-data-preparation/chapter1-ingestion-and-storage/section3-storage-ebs-efs-fsx - Transform · Data querying and wrangling stage
part1-data-preparation/chapter2-data-transformation - Querying · Data access and analytical querying
part1-data-preparation/chapter2-data-transformation/section1-querying - Wrangling · Cleaning and feature shaping
part1-data-preparation/chapter2-data-transformation/section2-wrangling
Part 2
Model preparation
- Model Prep · From prepared data to trained models
part2-model-preparation - Managed Services · AWS-managed ML building blocks
part2-model-preparation/chapter1-aws-managed-services - Model Train · Training concepts and SageMaker execution
part2-model-preparation/chapter2-model-training - Theories · Foundational modeling concepts
part2-model-preparation/chapter2-model-training/section1-modeling-theories - SageMaker Train · AWS managed training workflow
part2-model-preparation/chapter2-model-training/section2-sagemaker-training - GenAI · Generative AI as a separate modeling track
part2-model-preparation/chapter3-generative-ai
Part 3
Model deployment
- Deployment · Model serving and inference choices
part3-model-deployment - Docker EP · Self-managed container-based inference
part3-model-deployment/chapter1-docker-endpoint - SageMaker EP · AWS managed inference endpoint
part3-model-deployment/chapter2-sagemaker-endpoint
Part 4
Pipeline management
- Ops · Security and monitoring around the ML lifecycle
part4-pipeline-management - Security · Operational security boundaries
part4-pipeline-management/chapter1-security - Monitoring · Runtime and pipeline monitoring
part4-pipeline-management/chapter2-monitoring
Other components
- Course Notes · Repository overview for the AWS ML Associate learning path
README.md
Connections 28
- Course Notes → Data Prep: covers · Repository is organized into the first learning path
- Course Notes → Model Prep: covers · Repository is organized into the second learning path
- Course Notes → Deployment: covers · Repository is organized into the third learning path
- Course Notes → Ops: covers · Repository is organized into the fourth learning path
- Data Prep → Ingest & Store: includes · Data preparation begins with ingestion and storage
- Data Prep → Transform: includes · Data preparation continues with transformation
- Ingest & Store → Kinesis Ingest: entry · Streaming ingestion boundary
- Ingest & Store → S3 Storage: stores · Durable object storage foundation
- Ingest & Store → EBS/EFS/FSx: stores · Mounted/shared storage alternatives
- Transform → Querying: splits into · Transformation starts with querying
- Transform → Wrangling: splits into · Transformation continues with wrangling
- S3 Storage → Querying: feeds · Stored data is queried for downstream prep
- S3 Storage → Wrangling: feeds · Stored data is wrangled for ML use
- Model Prep → Managed Services: includes · Managed AWS services introduce platform abstractions
- Model Prep → Model Train: includes · Model preparation centers on training
- Model Prep → GenAI: includes · Generative AI is treated as a separate track
- Model Train → Theories: grounds · Training is paired with conceptual modeling foundations
- Model Train → SageMaker Train: executes · SageMaker is the managed training boundary
- Wrangling → Model Train: inputs · Prepared data flows into model training
- SageMaker Train → SageMaker EP: deploys to · Managed training leads into managed serving
- Deployment → Docker EP: includes · Deployment offers a containerized path
- Deployment → SageMaker EP: includes · Deployment offers a managed AWS path
- Ops → Security: includes · Security is a cross-cutting operational concern
- Ops → Monitoring: includes · Monitoring is a cross-cutting operational concern
- Security → Monitoring: paired with · Operational governance and visibility work together
- Security → Data Prep: governs · Security wraps ingestion and storage
- Monitoring → Model Prep: observes · Monitoring spans preparation and training
- Monitoring → Deployment: observes · Monitoring spans deployment and inference