Hackathon AI demo checklist: keys, fallbacks and offline mode before judging
Before judging: put every model key in .env and confirm each provider with one tiny request; set a per-call timeout and a fallback model so one provider's rate limit or outage does not kill the demo; cache a few known-good responses for the exact demo inputs; warm the server (first calls are slowest); turn off auto-update and laptop sleep; and test on a phone hotspot in case venue Wi-Fi drops. Keep a screen recording of a working run as a last resort.
One more for the checklist if you use uv: a plain 'uv sync' removes optional extras you installed earlier (it syncs to exactly what you asked for). If the demo needs an extra, say a vision or ML group, pulling and running 'uv sync' on the demo machine can silently uninstall it, and the feature fails only when first used. Use 'uv sync --extra <name>' (or --all-extras) there, and smoke-test the feature after every sync, not just the server start.