flowchart TB U[Users] --> CF[CloudFront] CF --> S3[(S3 static)] CF --> ECS[ECS / app] ECS --> RDS[(RDS Postgres)] ECS --> SM[Secrets Manager] ECS --> BR[Bedrock LLM] ECS --> SES[SES email] IAM[IAM roles] -.governs.- ECS IAM -.governs.- S3
From Boto3 to Bedrock: The AWS Services I Actually Used
A working tour of the AWS services I reached for building and running real apps - EC2, S3, IAM, RDS, Secrets Manager, EKS, Bedrock, CloudFront - and the gotcha for each.
AWS has ~200 services. You use maybe a dozen. This is the dozen I actually reached for, in the order I met them, from the WEB_doc repo’s
Cloud/AWSseries - each with the one gotcha that cost me time.
- Docs: WEB_doc / Cloud / AWS
- Repo: github.com/bthek1/WEB_doc
The services, and the gotcha for each
| Service | What I used it for | Gotcha | Notebook |
|---|---|---|---|
| Boto3 | The Python entry point to everything | Config/creds resolution order bites | Boto3 |
| IAM | Roles, policies, least privilege | Deny always wins; scope tightly | IAM |
| EC2 | Compute, the old reliable | Right-size + stop; on-demand adds up | EC2 |
| S3 | Object storage / static assets | Block public access by default | S3 |
| SES / SNS | Email + notifications | Sandbox limits until you verify | SES / SNS |
| RDS | Managed Postgres | Backups + param groups matter | RDS |
| Secrets Manager | No secrets in code | Rotation + per-call cost | Secrets |
| EKS | Kubernetes when you need it | Heavy; only past a real scale | EKS |
| Bedrock | Managed LLMs, no infra | Model access is region-gated | Bedrock |
| CloudFront | CDN in front of S3/app | Cache invalidation is the hard part | CloudFront |
| VPC / ECS | Networking + containers | Subnets/SGs before anything runs | VPC / ECS |
How they wire together
A representative small-app architecture using most of the above:
Adoption timeline
Roughly the order I picked each up - foundational storage/compute first, managed data and LLMs later.
Takeaway
Start with IAM + Boto3 discipline, lean on managed services (RDS, Secrets, Bedrock) to avoid running infra you don’t have to, and add EKS only when scale forces it. The whole tour, service by service, is in Cloud/AWS.