Machine Learning Lab

A testing ground for our data pipelines and ML models. We use real-world competitive gaming datasets as a sandbox for building and validating production-grade analytics infrastructure.

Each Paragon project targets a different data domain with unique challenges -- different data shapes, update frequencies, and prediction targets. The goal is to stress-test our tooling across varied real-time datasets before applying it to broader use cases.

Pipeline Architecture

01
Ingest
Pull from official and community APIs on configurable intervals
02
Transform
Normalize, deduplicate, and enrich raw data into analysis-ready schemas
03
Model
Run classification, regression, and clustering models against clean data
04
Serve
Expose results via REST API and real-time dashboards
Clash Royale
Live

Paragon Royale

Our first ML testing ground. Real-time match analytics and predictive modeling built on top of live competitive data from Clash Royale.

Visit Live Dashboard
Capabilities
  • Real-time data ingestion pipeline
  • Predictive win-rate modeling
  • Pattern recognition across match histories
  • Live dashboard with streaming updates
Marvel Rivals
In Development

Paragon Rivals

Expanding our pipeline to a new data domain. Hero-level statistical modeling, team composition analysis, and meta prediction.

Capabilities
  • Multi-variable classification models
  • Team composition clustering
  • Meta shift detection algorithms
  • Cross-match trend analysis
Overwatch 2
Planned

Paragon Overwatch

Next target dataset for our analytics pipeline. High-dimensional performance tracking across heroes, maps, and skill tiers.

Capabilities
  • High-dimensional feature extraction
  • Skill rating trend prediction
  • Composition synergy scoring
  • Session-level performance modeling
CS2
Planned

Paragon Strike

Applying our pipeline to FPS data. Round-by-round economy modeling, spatial aim analysis, and map-specific performance metrics.

Capabilities
  • Economy state prediction
  • Spatial heatmap generation
  • Round outcome classification
  • Player performance fingerprinting