Use gpt-4o-mini for prototypes
Use the verified deployment by default. Do not switch models unless tutors announce a shared deployment for the event.
Live from the admin dashboard
Team Costs
Per-team Azure spend snapshot from the operations dashboard. Auto-refreshes every 60 seconds. The full breakdown — tutor view, burn-down chart, annual context — is on the admin dashboard (sign-in required).
Current month spend
—
Month-end forecast
—
| Team | Status | Cap (KRW) | Used (KRW) | Used % |
|---|---|---|---|---|
| Loading live data… | ||||
Last refresh: — · Source: https://koica-tiu-azure.halla.ai/api/public/team-costs · Refreshes every 60 s
Six Optimization Rules
Six choices that keep the shared budget healthy for the full year. None of them weaken the demo.
For service-by-service unit prices and excluded technologies, see the technology and cost matrix.
Keep prompts under 500 tokens
System prompts run on every call. Long system prompts compound — trim ruthlessly during iteration.
Cap max_tokens at 400
Lab and demo workflows rarely need longer answers. JSON responses fit comfortably in 300-400.
Stay in mock mode while building UI
The sample template returns deterministic mock responses with no Azure calls. Switch to live only when the UI is settled.
Stop App Service / Container Apps after demo
Container Apps with --min-replicas 0 is free when idle. App Service B1 is not — stop it explicitly.
No background polling loops
A forgotten retry script can drain a team budget in minutes. Ctrl-C every background task before lunch.
Deploy walkthrough
CV bot end-to-end deploy with worked cost table (~₩900 per team).
Tech & cost matrix
Allowed services, excluded tools, and service-level cost estimates.
Azure Portal guide
Student sign-in, resource checks, screenshots, and safe Portal use.
Azure Lab + Credit Safety
Student rules, Foundry quickstart, CostSimulator, troubleshooting.