Behind the Scenes: Building “Captain Cool” — The Agentic Multi-Agent IPL Strategist Powered by Gemini
In the Indian Premier League (IPL), a split-second decision can define the fate of a multi-million dollar franchise. Should a leg-spinner bowl the 18th over against a left-handed pinch-hitter when the dew is setting in? Should the captain invoke the “Impact Player” rule right now, or hold it back?
Traditionally, captains rely on gut instinct, and analysts rely on static historical databases. But what if you could combine Dhoni’s calm strategic foresight, Ricky Ponting’s aggressive counter-analyses, real-time live match state integration, and advanced generative AI to create the ultimate virtual captain?
Welcome to Captain Cool — a state-of-the-art, multi-agent AI cricket strategist built using the Google Gemini stack and Next.js.
🌟 The Vision: Why “Captain Cool” Stands Out
Most hackathon entries build either simple wrappers around LLMs or basic static data dashboards. Captain Cool breaks this mold by introducing Collaborative Agentic Reasoning directly into live sports analytics.
Instead of asking a single AI to spit out a recommendation, Captain Cool simulates a high-stakes virtual dugout debate between specialized AI agents before presenting a unified, bulletproof tactic:
- The Stats Analyst (Data & Environment): Feeds live match API data and dynamically queries real-time weather and stadium dew conditions.
- The Strategist (Dhoni Style): Formulates the calm, optimal tactical decision, prioritizing matchups and situational math.
- The Devil’s Advocate (Ricky Ponting Style): Forcefully challenges the plan, highlighting critical flaws, matchup anomalies, and overlooked risks.
- Captain Cool (Consensus Synthesizer): The final decision-maker. It digests the debate, resolves the conflicts, and delivers a structured, authoritative captain’s call.
🏗️ Technical Architecture: Under the Hood
The application is built on a highly performant, responsive stack designed to deliver instantaneous, real-time streamed responses.
graph TD
A[User / CricAPI Live Data] -->|Match Context| B[Next.js API Route]
B -->|Fetch Weather| C[Open-Meteo API]
C -->|Dew & Wind Data| D[Agent 1: Stats Analyst]
D -->|Situation Brief| E[Agent 2: Strategist]
E -->|DHONI-style Tactic| F[Agent 3: Devil's Advocate]
F -->|PONTING-style Critique| G[Agent 4: Captain Cool]
G -->|Consensus Synthesis| H[SSE Streaming Stream]
H -->|Premium Glassmorphic UI| I[Client Dashboard]
1. The Collaborative Multi-Agent Loop
The core intelligence runs inside a Next.js API route (src/app/api/captain/route.js). It leverages the new @google/genai SDK and utilizes gemini-2.5-flash for lighting-fast generation speeds.
Here is how the sequential context handoff works:
// Agent 1: Stats Analyst queries stadium conditions
const analystResponse = await chat.sendMessage({
message: `Analyze this match state and check the weather for Chennai...`
});
// Agent 2: Strategist proposes the initial Dhoni-style plan
const strategistPlan = await ai.models.generateContent({
model: 'gemini-2.5-flash',
config: { systemInstruction: "You are MS Dhoni, the master tactician..." },
contents: `Stats: ${analystResponse}. Match Context: ${matchContext}`
});
// Agent 3: Devil's Advocate aggressively critiques
const devilCritique = await ai.models.generateContent({
model: 'gemini-2.5-flash',
config: { systemInstruction: "You are the Devil's Advocate (Ricky Ponting style)..." },
contents: `Strategist proposed: ${strategistPlan}`
});
// Agent 4: Captain Cool synthesizes into a 3-part consensus
const finalDecision = await ai.models.generateContent({
model: 'gemini-2.5-flash',
config: { systemInstruction: "Synthesize this debate into the final captain's call..." },
contents: `Strategist: ${strategistPlan}. Devil's Advocate: ${devilCritique}`
});
2. Solving the “Live Data” Challenge
Unlike static simulators, Captain Cool connects to CricAPI to fetch actual live match data.
Because live matches aren’t always active during development hours, we engineered a brilliant three-step lookup pipeline in src/app/api/live-match/route.js:
- Search specifically for the active IPL Series (
/v1/series?search=IPL). - Resolve the latest match from the tournament schedule (
/v1/series_info). - Fetch detailed innings-by-innings scorelines (
/v1/match_info) to build a perfect, up-to-the-minute tactical dataset for our agent debate loop.
3. Server-Sent Events (SSE) Streaming
To make the dashboard feel alive, we didn’t want the user waiting 15 seconds for all four agents to finish talking. We built a custom ReadableStream returning text/event-stream. As each agent finishes their thought, it is streamed immediately to the user’s dashboard, visualizing the internal dugout debate in real-time.
🎨 Premium Sports-Themed Glassmorphism UI
A great algorithm deserves an equally stunning interface. We rejected boring default templates in favor of a custom, highly premium design:
- Atmospheric Stadium Lights: A dynamic background featuring slowly pulsing radial gradients simulating Chennai/Wankhede floodlights under the night sky.
-
Premium Glassmorphic Panels: High-blur backdrops (
backdrop-filter: blur(16px)), bright top-left borders for 3D depth, and clean micro-animations. -
Color-Coded Agent Identities:
- 🔵 Stats Analyst: Cool Data Blue
- 🟢 Strategist: Calm Matchup Green
- 🔴 Devil’s Advocate: Aggressive Dissenting Red
- 🟡 Captain Cool: Golden Masterpiece Glow
📊 The Ultimate Output: What the Submission Replies With
When you hit the “Generate Captain’s Strategy” button, the debate wraps up, and Captain Cool delivers a distinct, beautifully structured three-part response:
1. The Next Decision
Tactic: Bring in the Left-Arm orthodox spinner from the Pavilion End. Immediately transition to a defensive 6-3 field setup on the leg side (deep mid-wicket, deep square leg, long-on back). Keep the Impact Player (pinch-hitting opener) padded up in the dugout.
2. The Reasoning
“Look, the data shows the dew is settling in fast, making the ball like a cake of soap. However, we’ve got a fresh right-handed batter who struggles immensely when forced to hit against the spin on a gripping pitch. Bringing the spinner now forces him to hit into the wind toward the longer boundary. It’s about squeezing the run rate, not just chasing wickets.”
3. What the Dissenting Agent Said
“The Devil’s Advocate strongly argued that exposing a spinner during heavy dew is suicide, as wet balls lead to full tosses and easy boundary balls. They suggested sticking to hard-length pace bowling. However, we have overridden this warning because the batter’s negative matchup index against left-arm spin (-35%) outweighs the wet ball risk if we instruct the bowler to bowl slightly flatter through the air.”
🚀 Impact & Future Scope
Captain Cool is more than a hackathon proof-of-concept. It represents the future of real-time sports broadcasting and tactical coaching:
- Interactive Live Broadcasts: Imagine streaming platforms (like JioCinema or Hotstar) letting viewers click a button to see what “Captain Cool” would do right now during a live match.
- Smart Coaching Systems: Local clubs and academy teams can input their current scorecard to receive elite, professional-level tactical debates on how to turn the game around.
- Voice-Activated Strategy: Integrating speech-to-text to let a physical captain discuss scenarios with the dugout assistant on the field.
🛠️ How to Run Locally
Get the dugout running on your machine in under 3 minutes:
# 1. Clone the repository
git clone https://github.com/your-username/captain-cool.git
cd captain-cool
# 2. Install dependencies
npm install
# 3. Create your .env.local file
CRIC_API_KEY=your_cricapi_key
GEMINI_API_KEY=your_gemini_api_key
# 4. Spin up the development server
npm run dev
Open http://localhost:3000 and start captaining your team!
Developed with ❤️ for the Google Gemini Developer Hackathon. Elevate your sports strategy today with Captain Cool!
@gdgcmu #gdgcloudpune #
