DEV Community: Md. Shakhawat Hossain

STRATYX is a data-driven assistant coach for esports teams

Md. Shakhawat Hossain — Thu, 29 Jan 2026 17:11:21 +0000

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

GRID Live Events (WebSocket) → Event Ingestion → Statistical Validation → Causal Engine → React UI ↓ <500ms total latency
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.

Spooky AI Search: Where Magic Meets Machine Learning

Md. Shakhawat Hossain — Wed, 03 Dec 2025 17:57:35 +0000

The Next Evolution in AI-Powered Search Has Arrived

The Problem With Search Today
We've all been there. You search for something important—a research topic, exam prep materials, or just trying to understand a complex concept—and you're bombarded with an endless list of blue links. No context. No guidance. Just overwhelming information overload.
Traditional search engines treat you like a data miner, forcing you to dig through dozens of results, cross-reference sources, and piece together your own understanding. But what if search could be more than just retrieval? What if it could be intelligent, contextual, and even... enchanting?

Enter Spooky AI Search
Spooky AI Search isn't just another search engine. It's a reimagined experience that combines cutting-edge AI technology with thoughtful design to create something truly magical. Built with React, TypeScript, and powered by Google Gemini AI, this next-generation search platform transforms how you discover, understand, and interact with information.

What Makes It Revolutionary?
AI-Powered Intelligence at Your Fingertips
Every search doesn't just return links—it returns understanding. Google Gemini 1.5 Flash analyzes your query in real-time, providing comprehensive summaries that synthesize information from multiple sources. It's like having a research assistant who reads everything for you and delivers the key insights.
Related Questions That Think Ahead
The AI doesn't just answer your question; it anticipates your next ones. After each search, you'll see AI-generated follow-up questions that guide you deeper into the topic, creating a natural path of exploration and discovery.
Trust You Can Actually Trust
In an age of misinformation, Spooky AI Search implements sophisticated credibility scoring algorithms. Every result comes with trust indicators—High, Medium, or Low—plus special badges for academic and government sources. You'll know at a glance which sources deserve your attention.

Features That Transform Your Experience
📚 Study Mode: Your AI Study Partner
Preparing for exams just got exponentially easier. Upload your lecture slides or course PDFs (up to 50 pages), and watch as the AI:

Identifies 5-8 key topics from your materials
Conducts automated research on each topic
Generates comprehensive summaries
Curates the top 3 resources per topic with direct links
Handles batch research across all topics simultaneously

It's like having a dedicated tutor who pre-studies everything and creates a personalized study guide just for you.
🌳 Focus Timer: Gamification Meets Productivity
The innovative Pomodoro-style focus timer takes concentration to the next level. Set your study duration, and watch a beautiful tree grow on your screen as you work. But here's the twist: if you switch tabs or lose focus, your tree dies.
This isn't just a timer—it's a commitment device that tracks your progress, celebrates your achievements (with flowers when you complete sessions!), and maintains stats on your total focus time and trees grown. It's productivity gamified with a conscience.
🎵 Ambient Music: The Soundtrack to Deep Work
Sometimes you need the right atmosphere to focus. The integrated music player provides copyright-free meditation music with full volume control, animated visualizations, and a sleek glass-morphism design. Your preferences persist across sessions, creating your perfect study environment every time.
🎨 A Visual Experience That Delights
This isn't just functional—it's beautiful. The interface features:

An animated night sky background with twinkling stars
Five flying owls with realistic wing animations
Forty glowing fireflies leaving light trails
A detailed moon with visible craters
Smooth gradient effects and glass-morphism throughout
Staggered animations that feel organic and alive

Every interaction is designed to feel smooth, responsive, and satisfying.

Built on Modern Technology
The technical foundation is as impressive as the features:
Frontend Excellence: React 18 with TypeScript ensures type safety and robust component architecture, while TailwindCSS provides a utility-first styling approach for rapid, consistent design.
Backend Power: Netlify Edge Functions offer serverless scalability, and Supabase provides a PostgreSQL database with pgvector for advanced vector operations.
AI Integration: Google Gemini AI delivers state-of-the-art natural language understanding and generation, while Google Programmable Search Engine provides comprehensive web results.
Performance Optimized: Every animation runs at 60 FPS, images load lazily, and React's efficient rendering ensures smooth interactions even with complex UI elements.

Real-World Impact
For Students
Transform your study sessions with AI-powered material analysis, focused work sessions that keep you accountable, and curated research that saves hours of manual searching.
For Researchers
Get instant summaries of complex topics, discover credible sources with verified trust indicators, and explore related questions that expand your research scope organically.
For Curious Minds
Enjoy a search experience that's actually enjoyable. The beautiful interface, helpful AI assistance, and thoughtful features make learning and discovery feel less like work and more like exploration.

Getting Started Is Easy
The platform is open-source and straightforward to deploy:

Clone the repository
Install dependencies with npm
Set up your API keys for Google Gemini and Search
Initialize Supabase for database functionality
Deploy to Netlify with a single command

Comprehensive documentation guides you through local setup scripts for both Windows and Unix systems, ensuring anyone can get up and running quickly.

The Future of Search
Spooky AI Search represents a fundamental shift in how we think about information retrieval. It's not about finding links—it's about gaining understanding. It's not about endless scrolling—it's about guided exploration. It's not about cold functionality—it's about delightful experiences.
This is search reimagined for the AI age, where technology doesn't just serve us information but helps us think, learn, and grow. The owls aren't just flying across your screen for aesthetics—they're symbols of wisdom, reminding us that great tools should inspire us to seek knowledge with curiosity and wonder.

**Try It Today
https://github.com/shakhawathossain07/Spooky-AI-Search

**Whether you're a student cramming for finals, a researcher diving into new topics, or simply someone who appreciates when technology is done thoughtfully, Spooky AI Search offers something special.
The code is open. The experience is magical. The future of search is here.
Ready to experience search differently?
Visit the repository, deploy your instance, and discover why the next generation of AI-powered search doesn't just find information—it illuminates understanding.

Built with React, TypeScript, Google Gemini AI, and a touch of magic. Licensed under MIT.

AI EXAM EVALUATOR With Intelligent Feedback

Md. Shakhawat Hossain — Thu, 11 Sep 2025 14:17:30 +0000

Over the past decades, the education system has remained identical, and this sector lacks innovation. One of the main problems in today's education system is that the manual evaluation of exam answer scripts and grading is biased, resulting in unfair marks for students. On the other hand, manually checking every exam paper for faculty members is time-consuming and inconsistent, and it lacks transparent feedback. That's why we figured out a solution—AI Exam Evaluator, your AI-powered grading assistance. From IELTS to A-Level, our system can evaluate handwritten exam answer scripts with the highest precision. Divide different questions and answers into cells, give marks and overall grades with transparent feedback, and include Human-on-the-loop approach so faculty members can work alongside AI. This makes our exam script evaluation system more reliable and trustworthy. Ready to transform the exam evaluation system? Try the AI Exam Evaluator now.

AI Exam Evaluator built with Kiro's intelligent assistance. Our development started with vibe coding with Kiro. First, we told Kiro what we wanted to create. Kiro gave me some ideas based on our requirements. We refined that through expressing our opinions and back-and-forth conversations. What followed was pure magic. Kiro completed the job instantly, a task that typically takes developers months to accomplish. The breakthrough came with Kiro's specification-driven approach and agent steering documents. Instead of an unplanned conversation, we created structured documentation divided into requirements, design, and small task lists (just like divide and conquer), which Kiro transformed into production-ready code. We also used prompt engineering techniques in the Vibe coding conversation with Kiro, and Kiro completed those better than expected. The most impressive moment was when Kiro generated thousands of lines of production-ready code in just minutes. Kiro didn't just write the code; they also set up automatic testing and a code quality analyzer, security vulnerability scanner, and documentation updater. But here's where it gets revolutionary. Kiro's agent has automated my entire development workflow. Every time we change something, Agent Hooks automatically updates the document, testing, and vulnerability checking. We also configured the MCP Server functionality to turn draft code into working software without bugs. The result—a production-ready AI exam evaluator ready to serve thousands of faculty members & students, built in record time with enterprise-grade security and performance.

This is the future of development. This is building with Kiro.

Try it out: AI Exam Evaluator
GitHub Repository