RoyalePro

RoyalePro is a Clash Royale deck analysis app I built to answer one practical question:

"If I run this deck in ladder right now, what should I expect and what should I change first?"

Live app: https://royalepro.vercel.app

What it does

• Build a deck with current Evo / Hero / Champion slot rules.
• Run analysis for archetype fit, role balance, matchup risk, and tower troop synergy.
• Simulate performance against common meta-style presets.
• Generate ML-assisted swap suggestions with predicted win-rate impact.
• Capture real match feedback (I Won / I Lost) to improve future model training.

Why this project is interesting

This is not just a UI project. It is a full decision system with multiple services:

• A frontend optimized for quick iteration while deck testing.
• A deterministic analysis layer for explainable scoring.
• A separate ML service for prediction and recommendation.
• A Postgres feedback pipeline for continuous improvement.

The hard part is balancing explainability and predictive quality without making the output feel like a black box.

Architecture

• Frontend: HTML, CSS, JavaScript
• API layer: Node.js serverless routes
• Rules/analytics backend: ASP.NET Core (C#)
• ML service: FastAPI + scikit-learn (Python)
• Data store: PostgreSQL (Supabase)

ML loop (how learning happens)

1. User analyzes a deck.
2. ML service returns predicted win rate + suggestions.
3. User submits match outcome (I Won / I Lost) with optional context (crowns, opponent archetype, trophies).
4. Feedback is stored in Postgres.
5. Training job uses synthetic + meta-prior + real feedback data to improve the model.

Engineering focus areas

• Input validation and schema constraints for cleaner training data.
• Meta-aware feature engineering (deck similarity to known strong lists).
• Defensive fallbacks when ML service is unavailable.
• Practical observability via learning status metrics and recent event history.

Security and operational notes

• No API secrets are committed to the repository.
• Runtime secrets are injected via environment variables.
• Database access is isolated behind backend services.
• Service endpoints are designed to fail safely when dependencies are offline.
• Vercel cron keeps the Render ML service warm at /api/ml/warm on a daily schedule (Hobby plan compatible).
• Set CRON_SECRET on Vercel so only authorized cron calls can hit warmup route.