omfgg-dev / hackathon_summary.md
therawr's picture
Initial development setup
8dce2a8
|
Raw History Blame Contribute Delete
5.14 kB

A newer version of the Gradio SDK is available: 6.32.0

Upgrade

Project Summary: Mad-Lib Generated Micro-Games (AI Hackathon Version)

🕹️ Overview

We’re building a Mad-Lib–style game generator where the user fills in a few playful blanks, and our AI turns those words into a fully playable micro‑game in seconds. The experience should be funny, surprising, and instantly shareable.

This approach supports rapid creativity, a controlled input schema, and parallel AI generation — perfect for an AI Hackathon.

Working Title OMFGG Which we will be vague about what it stands for, and randomize it. Examples:

Our Mad-Lib Factory Generates Games One Mistyped Form: Great Game! Obliviously Mashing Fields: Good Game Overloaded Machine Forming Goofy Games


🎯 AI Hackathon Requirements (Track 2: MCP in Action)

Event: MCP's 1st Birthday Hackathon Dates: November 14-30, 2025 (17 days) Track: Track 2 - MCP in Action Category: Consumer MCP Servers (best fit for our game generator)

Must-Haves for Submission

  • Build a Gradio app hosted as a HuggingFace Space
  • Demonstrate MCP (Model Context Protocol) usage
  • Show autonomous Agent behavior: planning, reasoning, execution
  • Must be original work created during Nov 14-30, 2025
  • Required deliverables:
    • HuggingFace Space with appropriate track tag in README
    • 1-5 minute demo video showing the app in action
    • Social media post about the project (link included in README)
    • Complete documentation in README
  • Mobile-friendly UI is strongly encouraged

Prize Structure (Track 2)

  • 1st Place (per category): $2,500 USD
  • 2nd Place: $1,000 USD
  • 3rd Place: $500 USD
  • Community Choice Award: $1,000 USD (based on engagement)
  • Sponsor Choice Awards: Various
  • API Credits: See below

Available Sponsor Credits

Our complete credit stack:

  • OpenAI: $25
  • Anthropic: $10
  • HuggingFace: $25
  • Modal: $250 (compute/deployment)
  • Blaxel: $250
  • ElevenLabs Creator: $44 (audio/TTS)
  • SambaNova: $25
  • Nebius Token Factory: $50

Allowed Tools & Resources

  • Vibe coding tools (Bolt, Lovable, Cursor) are permitted
  • Existing libraries and frameworks allowed
  • Core contributions must be original
  • Parallel agent workflows encouraged (perfect for our use case!)
  • Multimodal features encouraged

🧩 How We Define & Generate the Game

1. Adaptive Mad-Lib Blueprint

The user first selects a game mode, which determines the field set:

Game Modes:

  • Relaxing - Calm, meditative experiences
  • Funny - Absurd, humorous scenarios
  • Chaotic - Fast-paced, wild gameplay
  • Challenge - Goal-oriented with obstacles
  • Surprise Me - AI picks everything with minimal input

Based on the mode, the user fills in 5–8 adaptive fields:

Mode: Relaxing

  • Subject (what you interact with)
  • Vibe (calm/zen/cozy)
  • Setting (peaceful location)
  • Wildcard (gentle mechanic)

Mode: Funny

  • Subject (silly character/thing)
  • Action (absurd verb)
  • Vibe (comedic tone)
  • Setting (weird location)
  • Twist (unexpected element)

Mode: Chaotic

  • Subject (fast-moving character)
  • Action (frantic verb)
  • Obstacle (hazard/challenge)
  • Chaos Modifier (randomness)
  • Setting (dynamic location)

Mode: Challenge

  • Subject (player character)
  • Goal (win condition)
  • Obstacle (challenge/antagonist)
  • Setting (game arena)
  • Twist (power-up/mechanic)

Mode: Surprise Me

  • Vibe (just one input)
  • Everything else auto-generated

These structured inputs constrain creativity while maximizing fun and personalization.


2. Parallel Sub-Agents

Each field feeds a specialized agent (agents adapt based on mode):

Field Sub-Agent Output
Subject Character-Agent Traits, sprite/theme, or abstract element
Action/Goal Mechanic-Agent Interaction style, win condition (if any)
Vibe Style-Agent Colors, mood, tone, aesthetic
Obstacle Conflict-Agent Challenge logic (or absence of conflict)
Setting Level-Agent Background, layout, environment
Wildcard/Twist/Chaos Twist-Agent Special mechanics, power-ups, randomness

Agents run in parallel for speed and incorporate mode-specific instructions.


3. Composer Agent

The Composer Agent:

  • Collects all sub-agent outputs
  • Validates coherence
  • Constructs the final GameDef JSON
  • Saves it via MCP (e.g., Supabase)
  • Returns a shareable slug
  • Triggers the playable micro‑game in Gradio

📱 Game Characteristics

We aim for micro-games that are:

  • Quick to play (5–30 seconds)
  • Funny and chaotic
  • Mobile-first, vertical layout
  • Shareable and remixable
  • Generated in a few seconds
  • Simple but surprising

💥 Why This Works

  • Mad Libs create controlled randomness
  • Sub-agents allow parallel creativity
  • MCP makes tool integration modular
  • The system is fast enough for live demoing
  • Results feel personal and chaotic in a good way

🚀 Goal

Deliver a “What Did I Just Make?” experience:

Fill in a few words → AI generates a ridiculous micro‑game → Share with friends.