STRATYX: AI & data-driven assistant coach for esports

An AI & data-driven assistant coach for esports. It uses official GRID APIs to pull real match data and turn it into actionable coaching insights inside a single React dashboard.

Comment

Inspiration

Esports has exploded into a multi-billion dollar industry, yet coaching tools remain surprisingly primitive. While traditional sports have embraced data analytics for decades—think Moneyball revolutionizing baseball—esports coaches still rely heavily on intuition, manual VOD reviews, and scattered spreadsheets. We asked ourselves: What if an assistant coach could process thousands of in-game events in real-time, identify causal patterns invisible to the human eye, and deliver statistically-validated insights during live matches?

We were inspired by the untapped potential of GRID's official esports data APIs and the transformative power of AI to democratize elite-level coaching. Our mission was clear: build the Moneyball of esports—a data-driven assistant coach that empowers teams at every level to compete like world champions.

What it does

STRATYX is an AI-powered assistant coach that transforms raw match data into actionable coaching intelligence in under 500ms.

Core Features: 🎯 Real-Time Win Probability Engine — Bayesian estimation with Monte Carlo simulations (1000+ iterations) providing live win probability with confidence intervals and factor decomposition (economy, objectives, man-advantage, strategy debt)

🔬 Causal Inference System — Connects micro-actions (missed shots, poor positioning) to macro-outcomes (round losses, map defeats) through temporal graph analysis with statistically-validated causal weights

📊 Strategy Debt™ Metric — Our proprietary quantification of accumulated tactical disadvantages across game phases, helping coaches identify which bad habits are actually costing wins

🧠 ARIA - AI Assistant Coach — Powered by Gemini AI, provides natural language coaching advice, real-time tactical suggestions, and personalized improvement plans contextual to the live match

📈 Pattern Recognition — Detects recurring mistakes, success sequences, and phase-specific vulnerabilities using sliding-window analysis with minimum confidence thresholds

🎮 Multi-Game Support — Valorant, League of Legends, CS2, and Dota2 support via GRID's official APIs

Views: Live Dashboard — Real-time match intelligence with win probability charts, causal graphs, and priority alerts Coach Insights — Deep-dive analysis with AI-generated recommendations Player Analysis — Individual performance cards with impact scores, risk levels, and improvement plans How we built it Architecture Tech Stack: Frontend: React 18, TypeScript, Vite, Tailwind CSS Visualization: Recharts for data viz, Three.js for 3D AI assistant panel Data Layer: Apollo Client for GraphQL, WebSocket for real-time events AI: Google Gemini API for conversational coaching Statistics: Custom implementations of Chi-Square, Mann-Whitney U, Pearson correlation, and 95% confidence intervals Data Sources: GRID Central Data GraphQL API (historical match data) GRID Series State GraphQL API (live match state) GRID File Download API (event logs and end-state data) Key Technical Implementations: Temporal Feature Store — Time-series storage with player-specific indexing for sub-100ms feature retrieval Real-Time Sync Service — WebSocket primary with GraphQL polling fallback, auto-reconnect, and backpressure handling Statistical Validation Layer — Every insight requires p < 0.05 significance and includes effect size calculations Counterfactual Simulator — "What-if" scenario analysis for strategic planning

Challenges we ran into

🔄 Real-Time Processing at Scale — Achieving <500ms end-to-end latency while running statistical validation on every insight was incredibly challenging. We implemented aggressive event batching, priority queues, and sliding-window pattern analysis to meet our performance targets.

📊 Statistical Rigor vs. Speed — Running Chi-Square tests and calculating confidence intervals in real-time seemed contradictory. We solved this by pre-computing statistical thresholds and using incremental update algorithms.

🔗 Causal Inference in Non-Deterministic Games — Esports have high variance. A "bad" play can still work, and vice versa. We addressed this by building confidence-weighted causal graphs that accumulate evidence over time rather than making snap judgments.

🎮 Multi-Game Normalization — Valorant's agent abilities, LoL's lanes, and CS2's economy are fundamentally different. Creating a unified data model that preserves game-specific nuances while enabling cross-game patterns was a significant engineering challenge.

🔐 API Authentication & Rate Limiting — Managing WebSocket connections with proper authentication while gracefully falling back to polling during disconnects required careful state management.

Accomplishments that we're proud of

✅ <500ms Processing Latency — From raw event to validated insight in under half a second, guaranteed

✅ Statistical Rigor — Every single insight includes p-values, confidence intervals, and effect sizes. No more "gut feel" recommendations

✅ Strategy Debt™ — We invented a novel metric that quantifies accumulated tactical mistakes across game phases, something no existing tool provides

✅ Production-Ready Architecture — Auto-reconnecting WebSockets, fallback polling, data quality scoring, and comprehensive error handling

✅ ARIA AI Coach — A contextually-aware conversational AI that understands the current match state and provides personalized coaching

✅ Full GRID Integration — Successfully integrated all three GRID APIs (Central Data, Series State, File Download) with proper authentication

✅ Beautiful UX — Clean, responsive dashboard with real-time updates, 3D visualizations, and intuitive navigation

What we learned

📈 Statistics Matter — Early prototypes generated insights that "felt" right but couldn't withstand scrutiny. Adding statistical validation transformed the product from a toy to a tool professionals would trust.

⚡ Real-Time is Hard — The gap between "fast" and "real-time" is enormous. Every millisecond matters when coaches need information before the next round starts.

🎮 Domain Expertise is Critical — Building for esports required deep understanding of game mechanics, meta evolution, and what coaches actually need (not what we thought they needed).

🔬 Causal ≠ Correlation — Showing that a player dies often near B site is useless. Showing why they die (positioning, timing, utility usage) and connecting it to round outcomes is actionable coaching.

🤖 AI Augmentation > AI Replacement — Coaches don't want to be replaced by AI. They want tools that make them faster and more insightful. ARIA works with coaches, not instead of them.

What's next for STRATYX: AI & data-driven assistant coach for esports

Short-Term Roadmap: 🎥 VOD Integration — Sync insights with match recordings for visual review sessions 📱 Mobile Companion App — Coaching insights on the sidelines during LAN events 🗣️ Voice Commands — "ARIA, what should we focus on next round?" during live matches Medium-Term: 🏆 Tournament Mode — Track patterns across entire tournament brackets, identify meta shifts in real-time 🧬 Opponent Scouting — Automated analysis of opponent tendencies from historical data 📊 Custom Metrics Builder — Let coaches define their own KPIs and track them automatically Long-Term Vision: 🌐 API Platform — Open STRATYX insights to third-party tools and team management systems 🎓 Academy Integration — Connect amateur players with professional training methodologies 🏅 League Partnerships — Direct integration with esports leagues for real-time broadcast analytics

We believe STRATYX represents the future of esports coaching—where data science and AI work alongside human expertise to unlock performance that neither could achieve alone. The $1.8B esports industry deserves world-class coaching tools. STRATYX delivers.

Built With

Share this project:

Updates