GLAIZE

Sports have rules. The internet didn't—until now. Glaize is the answer: the first AI-powered fantasy league that turns the chaos of the creator economy into a formalized competitive sport.

Comment

Inspiration

I'm a creator myself, and I've formerly championed by favorite creators on the internet. I noticed something obvious but overlooked: The Creator Economy is already a sport. It has everything a professional league has:

Stats: Views, retention, CTR, subscriber growth. Leaderboards: The Trending Tab, Social Blade rankings. Intense Competition: 50M+ creators battling daily for a finite resource—attention. Implicit Rules: "Don't post during school hours," "Thumbnails must pop," "Drama spikes engagement." But while the creators were playing a high-stakes game, the fans were just spectators.

I asked: Why hasn't anyone formalized this ruleset? What if we took the unspoken mechanics of the attention economy and built an actual game engine around them?

This led to our core question: What if fantasy sports—a $30B industry—collided with the creator economy - a $250 billion market?

Fantasy football gives fans "skin in the game." They don't just watch players; they study them, strategize around them, and feel genuine joy (or heartbreak) based on weekly performance. Glaize brings that same emotional investment to YouTube and Twitch. It transforms passive viewing into active competition, turning the "implicit rules" of the creator economy into explicit gameplay mechanics.

What it does

Glaize is a fantasy sports platform for YouTube & Twitch creators, powered by an advanced AI strategist.

Draft & Compete: Users join leagues, draft a team of creators (MrBeast, PewDiePie, Kai Cenat), and compete in weekly head-to-head matchups.

Glaize Intelligence (The AI Strategist): A continuously running AI agent that acts as your personal "Front Office." It scans the web for breaking news, analyzes creator performance trends, and proactively suggests strategic moves (e.g., "MrBeast just dropped a video—play 'Viral Spike' now!").

"Bot Mind" Opponents: Users can play against AI personas that don't just use random logic—they have distinct personalities and strategies, trash-talking and reacting to your moves in real-time based on the Glaize Intelligence feed.

Real-Time Scoring: Points are awarded based on actual performance (Views, Subscribers, Peak Viewers) pulled live from YouTube and Twitch APIs.

"Hype Deck" Strategy: Play Hype Cards: 18 strategic action cards across 4 departments: Growth: Viral Spike, Trend Setter, Moonshot Crisis: The Ratio, Apology Video, Redemption Arc Talent: Main Character, Beef, The Link-Up Strategy: Poach, Insider Info, Niche Authority

Attention Capture: A unique mechanic where users can submit any viral social media link (TikTok, X, Reddit) during a matchup. The AI analyzes the content's "Attention Value" and awards instant points.

How we built it

Architecture:

Frontend: Expo SDK 54 + React Native for iOS, Android, and Web from a single codebase Backend: Supabase (PostgreSQL + Edge Functions on Deno) Styling: NativeWind (Tailwind CSS for React Native) State: TanStack Query for server state management AI: Gemini 3 Flash with Grounding for real-time card validation

Gemini Integration Deep Dive: The core innovation is using Gemini as a "Talent Agency Intelligence Officer" that validates gameplay claims in real-time. We instruct the model to act as an impartial referee, validating claims like "This creator started beef with X" or "This video is trending on Twitter."

We use:

Gemini 3.0 Flash for rapid card validation (<2 sec response) Grounding with Google Search to verify real-world events Structured JSON output for deterministic parsing Exponential backoff retries for fresh events (uploads <30 min old)

Challenges we ran into

Platform Parity: YouTube and Twitch have fundamentally different metrics (views vs. watch hours, subscribers vs. followers). We solved this with a unified logarithmic cost formula and platform-specific weights to normalize scoring.

Timing-Sensitive Validation: When a player plays "Viral Spike" 30 seconds after a video uploads, Gemini can't find data yet. We implemented an exponential backoff retry system (30m → 1h → 2h) that schedules re-validation for fresh events.

Anti-Cheat for Attention Captures: Preventing users from claiming their own creator's uploads as "viral moments" required:

URL deduplication (global hash constraint) Official channel blocking (can't capture content from the creator themselves) Content fingerprinting (blocks reposts/screenshots) Per-creator cooldowns (same creator = 4h gap)

Snake Draft at Scale: Implementing fair snake drafts (pick order reverses each round) with bot auto-drafting for incomplete lobbies required careful database trigger design and real-time subscription handling.

Orchestrating AI Agents: Coordinating multiple Gemini instances (Strategist, Validator, Reporter) to work together without creating race conditions or blowing up our token budget was a major hurdle. We solved this by creating a shared "Intelligence Feed" in Postgres that all agents read from and write to.

Accomplishments that we're proud of

Autonomous AI Bots: We are proud of creating AI agents that can play a complex fantasy sports game autonomously. They draft, trade, and chat without human intervention, making every league feel alive even with fewer human players. Glaize Intelligence: The "AI Strategist" genuinely feels like playing against a human. It notices when you make a mistake, capitalizes on it, and sends you a notification about it. It turns a static fantasy app into a living, breathing competitive experience. "Attention Capture": We successfully built a system that lets users submit any viral moment from the web, have it analyzed by AI for sentiment and impact, and instantly rewarded in-game. It turns the entire internet into a game board. Cross-Platform Parity: Successfully normalizing data from YouTube and Twitch so that creators from different platforms can exist fairly in the same draft pool.

What we learned

LLMs as Referees: Gemini's grounding capability transforms AI from "assistant" to "authority." We can make bold gameplay claims verifiable, not just decorative.

The Power of Stakes: The public Wire feed—showing card plays, captures, and Gemini verdicts—creates social accountability that makes the game dramatically more engaging than private scoring.

Logarithmic Scaling is Essential: Linear scoring breaks at scale (MrBeast getting 100M views would shatter any leaderboard). Log-scale scoring keeps competition tight while still rewarding top-tier creators.

Platform APIs are Fragile: YouTube quota limits, Twitch rate buckets, and social media's hostility to extraction forced us to build robust fallback and caching systems.

First Principles > Feature Parity: We didn't try to clone DraftKings. We asked "What makes creator fandom different from sports fandom?" and built mechanics (Hype Cards, Attention Capture, AI validation) that are native to the attention economy.

What's next for GLAIZE

Expansion: Integrating Spotify (Artists) and TikTok (Short-form specific slots) to cover the full spectrum of the creator economy.

Live Streaming Mode: Real-time scoring during creator livestreams with in-match card plays.

Creator Partnerships: Verified creators can see their fantasy stats, engage with teams, and earn from in-game branding.

Season Championships: Cross-league tournaments with prize pools funded by entry fees.

Gemini 3 Pro for Complex Disputes: Tribunal system for disputed card plays, powered by Gemini's deeper reasoning capabilities.

Built With

Share this project:

Updates