The latest articles on DEV Community by Amit Sumit (@amit_uc_). https://dev.to/amit_uc_ https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F2833698%2F10fba614-aa18-4c12-9764-4d27f7a51618.jpg https://dev.to/amit_uc_ en Amit Sumit Sun, 17 May 2026 14:24:58 +0000 https://dev.to/amit_uc_/building-captaincool-nexus-an-autonomous-ipl-tactical-intelligence-system-with-google-gemini-adk-1b0b https://dev.to/amit_uc_/building-captaincool-nexus-an-autonomous-ipl-tactical-intelligence-system-with-google-gemini-adk-1b0b <h2> Introduction </h2> <p>Cricket captaincy is one of the most complex real-time decision systems in sports.</p> <p>Every over involves:</p> <ul> <li>probabilistic reasoning</li> <li>psychological pressure</li> <li>pitch behavior</li> <li>momentum shifts</li> <li>matchup analysis</li> <li>field geometry</li> <li>risk management</li> </ul> <p>Traditional AI chatbots fail here because captaincy is not a single-answer problem.</p> <p>It is a continuous tactical debate.</p> <p>So instead of building another “AI cricket assistant”, I built:</p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Feyb0ya2fuvqeep7dnlvt.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Feyb0ya2fuvqeep7dnlvt.png" alt="War zone" width="800" height="380"></a></p> <h1> CaptainCool Nexus </h1> <p>An autonomous AI-powered IPL tactical operating system built entirely on the Google Gemini ecosystem using:</p> <ul> <li>Gemini 2.5 Pro</li> <li>Gemini 2.5 Flash</li> <li>Google ADK</li> <li>Gemini Function Calling</li> <li>Gemini Vision</li> <li>Antigravity</li> <li>Multi-Agent Orchestration</li> </ul> <p>The goal was to create a system that thinks like an IPL war room.</p> <h1> The Core Idea </h1> <p>CaptainCool Nexus simulates elite cricket intelligence through autonomous Gemini agents that:</p> <ul> <li>debate tactics</li> <li>challenge assumptions</li> <li>retrieve tactical memory</li> <li>simulate alternate futures</li> <li>analyze pitch conditions</li> <li>generate field placements</li> <li>compute pressure states</li> <li>explain decisions like real commentators</li> </ul> <p>This is not a chatbot.</p> <p>It is a tactical intelligence operating system.</p> <h1> Why Multi-Agent Architecture? </h1> <p>Cricket captaincy cannot be solved reliably with a single prompt.</p> <p>A single model tends to:</p> <ul> <li>overfit to one strategy</li> <li>ignore tactical tradeoffs</li> <li>lack internal disagreement</li> <li>miss psychological context</li> </ul> <p>So I designed a true multi-agent architecture using Google ADK.</p> <p>Each agent has:</p> <ul> <li>independent prompts</li> <li>separate reasoning style</li> <li>separate tactical role</li> <li>contextual memory</li> <li>tool permissions</li> </ul> <h1> Gemini Agent System </h1> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fyacrn5msveb0extbne5t.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fyacrn5msveb0extbne5t.png" alt="Field" width="800" height="382"></a></p> <h2> Match Strategist </h2> <p>Proposes bowling changes, batting adjustments, and tactical plans.</p> <h2> Statistical Intelligence Agent </h2> <p>Analyzes:</p> <ul> <li>matchup history</li> <li>strike rates</li> <li>death-over patterns</li> <li>projected scoring</li> </ul> <h2> Conditions Analyst </h2> <p>Evaluates:</p> <ul> <li>dew</li> <li>pitch deterioration</li> <li>grip</li> <li>bounce</li> <li>turn probability</li> </ul> <h2> Momentum Psychologist </h2> <p>Tracks:</p> <ul> <li>pressure spikes</li> <li>emotional momentum</li> <li>collapse probability</li> <li>crowd pressure</li> </ul> <h2> Devil’s Advocate </h2> <p>Aggressively attacks tactical assumptions and forces strategic revisions.</p> <h2> Captain Persona Engine </h2> <p>Models tactical personalities of:</p> <ul> <li>Dhoni</li> <li>Rohit</li> <li>Kohli</li> <li>Gambhir</li> <li>Hardik</li> <li>Cummins</li> </ul> <h2> Tactical Simulator </h2> <p>Runs counterfactual simulations and alternate futures.</p> <h2> Commentary Narrator </h2> <p>Transforms AI reasoning into broadcast-style tactical commentary.</p> <h1> Google ADK Orchestration </h1> <p>The entire orchestration pipeline runs through Google ADK.</p> <p>The orchestrator:</p> <ol> <li>routes tactical context</li> <li>activates relevant agents</li> <li>manages debate rounds</li> <li>propagates confidence</li> <li>invokes tools</li> <li>retrieves memory</li> <li>synthesizes consensus</li> </ol> <p>This creates visible autonomous collaboration instead of fake “multi-agent roleplay”.</p> <h1> Gemini Function Calling </h1> <p>One major requirement was genuine tool usage.</p> <p>I implemented native Gemini function calling for:</p> <ul> <li>player matchup analytics</li> <li>pressure index calculations</li> <li>tactical simulations</li> <li>venue intelligence</li> <li>pitch analysis</li> <li>live match parsing</li> </ul> <p>The agents autonomously decide:</p> <ul> <li>when to invoke tools</li> <li>which tools to use</li> <li>how to incorporate outputs</li> </ul> <p>This made the reasoning process significantly more believable.</p> <h1> Tactical Debate Engine </h1> <p>The debate engine became the heart of the system.</p> <p>Instead of generating a single answer:</p> <ol> <li>Strategist proposes</li> <li>Stats Agent validates</li> <li>Conditions Analyst critiques</li> <li>Devil’s Advocate attacks</li> <li>Persona Engine modifies intent</li> <li>Tactical Simulator projects outcomes</li> <li>Consensus Engine finalizes decision</li> </ol> <p>The UI visualizes:</p> <ul> <li>confidence shifts</li> <li>rejected strategies</li> <li>tactical pruning</li> <li>consensus convergence</li> <li>risk propagation</li> </ul> <p>This makes the system feel alive.</p> <h1> Tactical Multiverse Engine </h1> <p>One of my favorite features was the Tactical Multiverse Engine.</p> <p>The system simulates alternate universes:</p> <ul> <li>aggressive bouncer attack</li> <li>spin introduction</li> <li>defensive wide-line strategy</li> <li>field compression</li> <li>matchup traps</li> </ul> <p>Each branch computes:</p> <ul> <li>win probability delta</li> <li>collapse risk</li> <li>momentum volatility</li> <li>tactical tradeoffs</li> </ul> <p>This transforms tactical analysis into a probabilistic simulation system.</p> <h1> Tactical Memory System </h1> <p>CaptainCool Nexus continuously remembers:</p> <ul> <li>failed plans</li> <li>successful overs</li> <li>pressure spikes</li> <li>batsman tendencies</li> <li>field outcomes</li> <li>captain behavior</li> </ul> <p>Agents reference memory dynamically during debates.</p> <p>Example:</p> <blockquote> <p>“The slower bouncer failed against Pant in Over 14.”</p> </blockquote> <p>This created innings continuity instead of isolated predictions.</p> <h1> Multimodal Intelligence </h1> <p>Gemini Vision powers:</p> <ul> <li>pitch image analysis</li> <li>scorecard screenshot parsing</li> <li>contextual image reasoning</li> </ul> <p>The system can infer:</p> <ul> <li>dryness</li> <li>crack density</li> <li>turn expectation</li> <li>dew absorption</li> <li>pace deterioration</li> </ul> <p>This made the platform feel like a real tactical intelligence system.</p> <h1> Observability &amp; Telemetry </h1> <p>One important goal was making AI reasoning observable.</p> <p>The observability dashboard tracks:</p> <ul> <li>orchestration cycles</li> <li>token throughput</li> <li>active agents</li> <li>confidence propagation</li> <li>memory retrieval</li> <li>live telemetry</li> <li>tool execution traces</li> </ul> <p>Most AI projects hide orchestration.</p> <p>CaptainCool Nexus exposes it.</p> <h1> Tech Stack </h1> <h2> Frontend </h2> <ul> <li>Next.js 15</li> <li>TypeScript</li> <li>TailwindCSS</li> <li>Framer Motion</li> <li>shadcn/ui</li> </ul> <h2> AI Stack </h2> <ul> <li>Gemini 2.5 Pro</li> <li>Gemini 2.5 Flash</li> <li>Google ADK</li> <li>Gemini Function Calling</li> <li>Gemini Vision</li> </ul> <h2> State &amp; Visualization </h2> <ul> <li>Zustand</li> <li>React Query</li> <li>Recharts</li> <li>React Flow</li> </ul> <h2> Deployment </h2> <ul> <li>Vercel</li> </ul> <h1> Biggest Engineering Challenge </h1> <p>The hardest challenge was balancing:</p> <ul> <li>believable tactical intelligence</li> <li>orchestration complexity</li> <li>UI responsiveness</li> <li>streaming reasoning</li> <li>explainability</li> </ul> <p>Without turning the system into fake “AI theater”.</p> <p>The solution was:</p> <ul> <li>visible reasoning evolution</li> <li>observable telemetry</li> <li>real tactical disagreement</li> <li>confidence propagation</li> <li>structured orchestration</li> </ul> <h1> What I Learned </h1> <p>This project completely changed how I think about AI systems.</p> <p>The future is not:</p> <blockquote> <p>“one super prompt”</p> </blockquote> <p>The future is:</p> <ul> <li>orchestration</li> <li>specialization</li> <li>memory</li> <li>simulations</li> <li>explainability</li> <li>autonomous collaboration</li> </ul> <p>Agentic systems become significantly more powerful when models can:</p> <ul> <li>disagree</li> <li>revise</li> <li>simulate</li> <li>remember</li> <li>critique</li> </ul> <h1> Final Thoughts </h1> <p>CaptainCool Nexus started as a hackathon project.</p> <p>But it evolved into something much larger:<br> an experiment in autonomous tactical intelligence.</p> <p>Cricket became the perfect medium for exploring:</p> <ul> <li>multi-agent reasoning</li> <li>strategic simulations</li> <li>memory systems</li> <li>probabilistic decision-making</li> <li>AI explainability</li> </ul> <p>And Google Gemini + ADK made that possible.</p> <h1> Links </h1> <h2> GitHub Repository </h2> <div class="ltag-github-readme-tag"> <div class="readme-overview"> <h2> <img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"> <a href="https://github.com/AmitChigare" rel="noopener noreferrer"> AmitChigare </a> / <a href="https://github.com/AmitChigare/captaincool-nexus" rel="noopener noreferrer"> captaincool-nexus </a> </h2> <h3> Autonomous IPL Tactical Intelligence System powered by Gemini Multi-Agent Orchestration </h3> </div> <div class="ltag-github-body"> <div id="readme" class="md"> <div class="markdown-heading"> <h1 class="heading-element">CaptainCool Nexus 🏏🤖</h1> </div> <p><strong>CaptainCool Nexus</strong> is a research-grade, autonomous AI cricket command center Built on a high-performance Next.js 15 App Router architecture, it simulates a state-of-the-art Formula 1 pit-wall and military command center, entirely dedicated to real-time cricket tactical intelligence.</p> <div class="markdown-heading"> <h2 class="heading-element">🚀 Tech Stack</h2> </div> <ul> <li> <strong>Framework:</strong> Next.js 15 (App Router)</li> <li> <strong>Language:</strong> TypeScript</li> <li> <strong>Styling:</strong> Tailwind CSS v4 (with custom Dark Tactical Theme)</li> <li> <strong>Animations:</strong> Framer Motion</li> <li> <strong>UI Components:</strong> shadcn/ui &amp; Lucide React icons</li> <li> <strong>State Management:</strong> Zustand</li> <li> <strong>Data Fetching:</strong> React Query</li> <li> <strong>AI Ecosystem integration:</strong> Gemini 2.5 Pro, Gemini 2.5 Flash, Vertex AI, Google ADK</li> </ul> <div class="markdown-heading"> <h2 class="heading-element">📂 Architecture Structure</h2> </div> <p>This repository is organized to support multi-agent collaboration, memory persistence, and telemetry:</p> <div class="snippet-clipboard-content notranslate position-relative overflow-auto"> <pre class="notranslate"><code>src/ ├── agents/ │ ├── orchestration/ # Coordinates Gemini 2.5 Pro and Flash │ ├── persona/ # "Captain" Persona (MS Dhoni style logic) │ └── consensus/ # Resolves tactical conflicts between agents ├── components/ │ ├── dashboard/ # Main F1 Pit-wall UI (CommandCenter.tsx) │ ├── panels/</code></pre>…</div> </div> </div> <div class="gh-btn-container"><a class="gh-btn" href="https://github.com/AmitChigare/captaincool-nexus" rel="noopener noreferrer">View on GitHub</a></div> </div> <h2> Screenshots </h2> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fhbbvjyuyhj7htq7jwawr.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fhbbvjyuyhj7htq7jwawr.png" alt="War Room" width="800" height="381"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F2jeedb83nuopgfi7f5v0.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F2jeedb83nuopgfi7f5v0.png" alt="Tactical Debate Engine" width="800" height="379"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fvpz8xc595kvu0fxzrrc2.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fvpz8xc595kvu0fxzrrc2.png" alt="Multimodal Intelligence" width="800" height="383"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fgc1fdobq81eqggtencba.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fgc1fdobq81eqggtencba.png" alt="Field Intelligence Engine" width="800" height="380"></a></p> <p>Built with Google Gemini, Google ADK, and Antigravity for Agentic Premier League by GDG Cloud Pune.</p> agents ai gemini showdev