Draft Board with Recommendations - Mid-draft with AI recommendations panel open, win probability gauge visible, some picks already locked
Lock-In Animation - Capture mid-animation (Shockwave or X-Strike), champion name visible with team color glow
Composition Analysis Panel - Both teams' damage profiles, CC ratings, scaling curves visible
AI Commentator Speaking - Panel showing narration text with the tone emoji, "LIVE" badge visible
Team Selection Lobby - Full-screen with both teams selected, animated VS divider, "START DRAFT" button glowing
Scouting Report - Full report showing target bans, danger picks, and one-trick alerts sections
Champion Browser - Grid view with filters, showing win rate color coding
Champion's Page - Performance Stats, Metics like Win Rate, Pick Rate, Ban Rate etc., Side Analysis, Best Synergies, Toughest Matchups,
Team Scouting Database - teams analyzed across series
About Page - Hero with stats bar (2,282 games, 56 teams, etc.)
Share & Download Draft/Scout Reports - Save as PDF, Image, Share on on X, Reddit, Copy for Discord, Share via other platforms

DraftMind AI

Inspiration

2 AM. Worlds 2023. Cloud9 vs T1.

I watched their coach stare at the screen. 27 seconds on the clock. Hand hovering over the pick. In that moment, he had to recall 10 hours of preparation - VOD reviews, spreadsheets, analyst calls - and make the right choice.

I thought: this is insane. All that data lives in documents. When the draft actually happens, coaches are alone with their memory.

What if they weren't?

The GRID API gave me 2,282 professional games from 56 teams. Real pro play, not Reddit theory. For the first time, I could build a draft companion grounded in what actually happens on stage.

DraftMind AI: data at the speed of competition.

What it does

Select two teams. Run the draft steps. Get AI recommendations at every decision.

Core Features:

AI Recommendations - Top 5 champions scored on meta strength, team affinity, counter matchups, and composition synergy. Confidence ratings show data quality.
Live Win Prediction - XGBoost ML model with 40 features. Watch the gauge shift with every pick. Temperature scaling keeps predictions realistic.
AI Commentator with Voice - Gemini generates analysis, Edge TTS speaks it. "Cloud9 locks Azir - this signals a scaling teamfight comp!" Hear insights while you think.
Lock-In Animations - Five dramatic variants (Shockwave, Vortex, Glitch, Eruption, X-Strike). Every pick feels like a statement.
Scouting Reports - Six sections: target bans, danger picks, one-trick alerts, draft tendencies, shared battleground. PDF export.
Composition Analysis - Damage profiles, CC ratings, scaling curves, synergy scores, archetype detection.
Export & Share - PDF, PNG, Twitter, Reddit, Discord. One click.

How we built it

Data: GRID API → 169 MB raw JSON → processed into champion stats, team profiles, synergy pairs, draft patterns. All in-memory for speed.

Intelligence: Four-signal recommendation engine (meta + team + counter + composition). XGBoost win predictor trained on real compositions. Hierarchical confidence based on sample size.

AI: Gemini 2.0 Flash with engineered prompts. Edge TTS with tone-aware modulation. Combined endpoint returns text + audio together - zero latency.

Frontend: React 18 + TypeScript + Tailwind + Framer Motion. Designed for pressure - recommendations appear instantly, win probability always visible.

Stack: Python, FastAPI, XGBoost, Gemini, Edge TTS, React, Vite, Cloud Run, JetBrains IDEs + Junie AI.

Challenges

Small-n problem: Pro teams have limited games. Solution: hierarchical intelligence - league baseline → team layer → head-to-head layer. Always show confidence.

30-second window: Coaches can't navigate dashboards under pressure. Solution: instant recommendations, single-number win probability, voice commentary so you can listen while thinking.

Real AI, not theatre: Easy to wrap an LLM around stats and call it "AI-powered." We refused. XGBoost is a real model. Recommendations have traceable signals. Gemini synthesizes, it doesn't hallucinate strategy.

Accomplishments

It works. Not a mockup. Run 20 draft steps, get recommendations that make competitive sense, watch win probability shift, hear the AI react.

The moment it clicked: I picked Azir, the gauge nudged blue. Gave away Nautilus, it swung back. The draft as a living game, quantified.

Voice changed everything. TTS was a last-minute add. Now it feels like a broadcast, not a spreadsheet.

Looks like a product. Animated lobby, dramatic lock-ins, spring animations, esports dark theme. Something Cloud9 would actually deploy.

What we learned

Data engineering is the competitive advantage. Everyone has GRID. Feature extraction is where value lives.
Design is how intelligence becomes usable. Brilliant engine + confusing UI = worthless under pressure.
AI needs constraints. Our narrator synthesizes real data, it doesn't make up strategy.
The best features are unplanned. Voice wasn't in the original spec.

What's next

VALORANT support - GRID has the data
Historical replay - What would DraftMind recommend in famous matches?
Broadcast integration - Win probability on official streams
The dream - Every esports team with AI draft intelligence at their fingertips