League Wrapped

Landing page with search form
Player Dashboard
Hidden gems, Head-to-head comparison with another player
Achievements, Stats, Champion performance, role distribution, playstyle DNA
Top 5 moments of 2025, Performance charts
Squad Synergy analysis
AI insights
Loding player data
Shareable cards

Inspiration

League of Legends has 150 million players who play hundreds of ranked games each season. Stat sites like OP.GG show KDA and win rates, but don't explain what makes each player's style different or how they've improved. I built League Wrapped to surface those insights: playstyle patterns, performance evolution, and standout games.

What it does

League Wrapped analyzes 100+ ranked matches and uses Amazon Bedrock to generate personalized AI insights. The platform combines data visualization (performance trends, champion analysis, role distribution) with 8 types of AI-generated content: season story, coaching feedback, playstyle personality, top moments commentary, and squad synergy analysis. It also detects 7 non-obvious patterns like time-of-day performance, tilt behavior, and comeback wins that stats sites miss.

The AI coaching is direct and honest. Instead of "good job, keep practicing," it says things like "Your vision score averages 18 per game—35% below your rank's median. Place 3-4 control wards near objectives" or "32% win rate on this champion over 25 games. Practice in normals first." Every insight includes specific numbers and rank-appropriate benchmarks.

Everything is shareable: downloadable cards for social media, one-click X posts, player-vs-player comparison mode.

How I built it

I'm using Amazon Bedrock's Nova Micro model ($0.00245 per user) for all AI insights. The tech stack is Next.js 15 + React 19 on the frontend, FastAPI + Pandas on the backend, deployed on AWS EC2 with Nginx and PM2. Data comes from Riot Games API; I fetch 100+ matches per player, process them into aggregate statistics with Pandas, then send that to Bedrock to generate 8 parallel insights in 3-4 seconds.

I also tried Amazon Polly for AI-narrated videos, but testing showed they just repeated UI content without adding value, so I removed the feature but kept the code.

Three optimizations keep costs low: Nova Micro saves 70-90% vs alternatives, pre-computing stats cuts token usage by 95%, and 24-hour caching reduces costs by 80% on repeat visits. At 1M users, total cost is $505/month. With 1% premium conversion at $4.99/month, that's $50K revenue = 99x ROI.

Challenges I ran into

Riot migrated from summoner names to Riot IDs (Name#TAG), breaking player lookups. I built automatic fallback that tries all common region tags if none is specified.

AI costs could spiral fast at scale. I solved this with Nova Micro (70-90% cheaper), pre-computation (95% token reduction), and 24-hour caching (80% cost savings).

The Polly video feature taught me that multi-modal formats need unique value. Videos just repeated UI content, so I removed the feature after testing. Shareable cards work because they enable social distribution; videos didn't offer that.

What I learned

Bedrock is production-ready. Nova Micro delivers quality at $0.00245 per analysis. Pre-computation cuts token usage by 90%, and 24-hour caching achieves 80% hit rates. Parallel processing (8 concurrent Bedrock calls) finishes in 3-4 seconds vs 20+ sequential.

The Polly experiment taught me that technical success doesn't guarantee UX value. Multi-modal features need to provide something unique, not duplicate what's already visible.

Players prefer honest, data-driven feedback over generic encouragement. Specific metrics with rank comparisons work better than vague advice.

What's next for League Wrapped

Immediate Priorities (Next 2 Months)

1. Video Export Feature (Reimagined)

Target: Social media sharing, premium monetization

Status: Initial Polly implementation removed after testing revealed UX issues
Next Iteration: Redesign to provide unique value beyond UI duplication
Amazon Polly: Text-to-speech narration with personalized insights not visible on screen
Video Generation: 30-60 second animated recap with exclusive content
Export Formats: MP4 optimized for Instagram Reels, TikTok, Twitter
AWS Stack: Lambda for video rendering, S3 for storage