multi-agent-lab / docs /strategy /hackathon-prize-map.md
agharsallah
feat: scaffold hackathon agent theater
42e32ed
|
Raw
History Blame Contribute Delete
1.45 kB

A newer version of the Gradio SDK is available: 6.19.0

Upgrade

Hackathon Prize Map

Primary Trail: Thousand Token Wood

Judging criteria and build response:

  • Genuinely delightful: optimize the first 15 seconds for surprise, humor, and visual polish.
  • AI is load-bearing: make every turn visibly generated by agents from ledger context.
  • Originality: frame the app as a tiny theater engine, not another chat UI.
  • Polish of Gradio app: custom theme, live state, examples, demo mode, and a strong opening seed.

Secondary Prizes

  • Community Choice: make the app instantly shareable, with funny generated moments and a short public demo clip.
  • OpenAI Track: keep a clean provider abstraction and document the small-model orchestration pattern.
  • Tiny Titan: maintain a <=4B mode and show how specialization compensates for model size.
  • Best Agent: expose the judge/conductor/agent loop clearly and make agent behavior inspectable.
  • Off-Brand UI: push Gradio past defaults with custom CSS and a stage-like layout.
  • Best Demo: prepare a concise story arc, recorded run, screenshots, and social post copy.
  • Judges' Wildcard: preserve the weirdness. Do not sand the concept down into a generic framework demo.

Demo Success Bar

Within 90 seconds, a judge should understand:

  1. This is a playful AI-native toy.
  2. Multiple small agents are doing different cognitive jobs.
  3. The ledger makes the system observable and resilient.
  4. The user can disturb the world and get surprising consequences.