gpt-4o-mini not available in koreacentral
Switch LOCATION to eastus, swedencentral, or japaneast and re-run from step 3b.
End-to-end deploy example
Deploy a CV Feedback Bot in 90 Minutes
One complete, replicable example from git clone to a live Azure URL. Built around the cheapest viable architecture: gpt-4o-mini + scale-to-zero Container Apps. Per-team cost stays under ₩900.
What You Build
A small FastAPI app
Single-page UI + /review endpoint. About 80 lines of Python.
Structured CV feedback
Strengths, weaknesses, suggestions, plus red-flag detection for private contact info.
Live on Azure
Container Apps with --min-replicas 0. Idle cost ₩0.
Cost Per Team Demo (measured)
Measured against an actual end-to-end deploy of the CV Feedback Bot. Assumes ~100 gpt-4o-mini calls (50 dev/test + 50 demo) and one active hour of Container Apps Consumption at 0.5 vCPU + 1 GiB. Two registry choices shown side-by-side.
| Component | Quantity | Unit | Ghcr.io path | ACR Basic path |
|---|---|---|---|---|
| Azure OpenAI gpt-4o-mini input | ~150K tokens (100 calls × 1.5K) | $0.000150 / 1K | ~₩30 | ~₩30 |
| Azure OpenAI gpt-4o-mini output | ~30K tokens (100 calls × 300) | $0.000600 / 1K | ~₩25 | ~₩25 |
| Container Apps compute | 1800 vCPU-sec + 3600 GiB-sec | within monthly free grant | ₩0 | ₩0 |
| Idle outside demo (scale-to-zero) | min-replicas = 0 | — | ₩0 | ₩0 |
| Container registry | Ghcr.io free / ACR Basic shared 5 teams · 1 mo | ACR $5 / 8 | ₩0 | ~₩843 |
| Per-team total | ~₩55 | ~₩900 | ||
Why two paths: GitHub Container Registry (ghcr.io) is free for public repos. Azure Container Registry Basic adds about ₩1,350/team/month when shared across 5 teams. Pick Ghcr.io for minimum cost; pick ACR if your team already has it provisioned by operations.
Container Apps free grant: every Azure subscription gets 180,000 vCPU-seconds + 360,000 GiB-seconds + 2,000,000 requests per month free. One demo hour at 0.5 vCPU + 1 GiB consumes only 1% of the vCPU grant, so compute is effectively ₩0 within these limits.
Eight Steps
High-level overview. Each card links to the matching section of the full walkthrough.
1
Clone the example branch
~2 min
git clone the template, checkout example/career-cv. No Azure cost yet.
2
Run locally in mock mode
~5 min
uvicorn + browser test. Submit a CV, verify the UI renders four feedback blocks. Still ₩0.
3
Create Azure resources
~6 min
az CLI: resource group, Azure OpenAI + gpt-4o-mini deployment, ACR, Container Apps Environment.
4
Connect locally to real Azure
~3 min
Paste endpoint + key into .env. Submit a CV — mode flips from "mock" to "azure".
5
Build and push the Docker image
~2 min
docker build + push to ACR. Or az acr build if Docker is not installed.
6
Deploy with scale-to-zero
~2 min
az containerapp create --min-replicas 0 --max-replicas 1. Idle cost stays at ₩0.
7
Verify the live URL
~1 min
curl /health and POST /review. Take a screenshot of the live UI for your demo backup.
8
Tear down
~5 min
az group delete. The single most important step for cost safety — do not skip.
When You Are Done
All four boxes ticked before you move on to polishing the slide deck:
- The live URL responds in under 5 seconds on a warm call.
- A demo screenshot is saved in /demo/screenshot.png.
- The team README explains: problem, user, Azure services used, cost-per-review.
- The repository has no .env, no API keys, no Azure passwords in commit history.
Watch Out
Cold start is 10+ seconds
Send a warm-up curl to /health one minute before the demo.
Forgot to tear down → bill keeps creeping
Set a calendar reminder for az group delete --name $RG --yes the day after the demo.
Ready to deploy?
The full walkthrough has every command, every expected output, and the teardown script. Bookmark it and follow top-to-bottom.