Cloud portability: AWS → Azure
Most of my cloud AI work has been on AWS — architecture evaluated and partly operated as part of a paid engagement building a sovereign consultation-analysis platform for a New Zealand qualitative-research firm (see the Stoa case study), plus the upstream AWS stack (ECS Fargate, Batch, SQS) credited to that project's origin and extended by me. This page maps that AWS-shaped experience onto the Azure AI stack, service for service.
Mapping
| Capability I shipped on AWS | AWS service | Azure equivalent |
|---|---|---|
| Managed LLM inference (Llama 70B, structured outputs) | Bedrock | Azure OpenAI / Foundry model catalog |
| Batch LLM pipeline jobs | AWS Batch + SQS | Container Apps Jobs + Storage Queues |
| App hosting | ECS Fargate | Container Apps |
| GPU self-hosting (Ollama) | EC2 p3-class | NC-series VMs |
| Secrets | Secrets Manager | Key Vault |
| Identity | IAM / Cognito | Entra ID |
| Serverless glue | Lambda + EventBridge | Functions + Event Grid |
I have not run Azure hands-on; this maps where each piece of my AWS work lands on the Azure AI stack.
Evidence
- Exhibit 02 — Bedrock/LangChain structured outputs
- Exhibit 10 — Terraform (IaC discipline; AWS stack credited as upstream, extended by me)
- Capabilities matrix — every claim on this page traced to evidence