Optimize AI Data Center Operations with Gemini 2.5 Pro & CrewAI
Inspired by the immense scale and cost of new AI data centers like Stargate, this challenge tasks developers with creating an intelligent, agentic system to optimize data center operations. Participants will build a CrewAI-based team of specialized agents, powered by Gemini 2.5 Pro, to manage resource allocation, predict maintenance needs, and enhance energy efficiency. This system will ingest simulated real-time telemetry data (power consumption, temperature, workload metrics) and make proactive operational decisions. The solution will emphasize hybrid instant/deep reasoning, allowing agents to react immediately to critical alerts while performing complex, multi-modal analysis (e.g., correlating thermal images with power logs) for long-term optimization strategies. Agents will leverage RAG against vast operational manuals and equipment specifications, and employ adaptive thinking budgets to intelligently manage their reasoning effort for different tasks, ultimately aiming to reduce operational expenditure and improve system uptime.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a CrewAI multi-agent system with Gemini 2.5 Pro agents for AI-driven data center optimization, using hybrid reasoning, RAG, and adaptive thinking budgets to manage resources and predict maintenance.
How work is evaluated
The solution will be evaluated on its ability to accurately identify and propose optimizations for simulated data center scenarios (e.g., energy spikes, potential hardware failures), its effective use of Gemini 2.5 Pro's capabilities, the quality of agent collaboration, and the demonstrable efficiency gains (simulated).
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Design and orchestrate a role-based multi-agent team using CrewAI for complex operational management.
Develop RAG strategies for agents to access and utilize extensive operational knowledge bases.
Apply adaptive thinking budgets to intelligently manage agent computational resources for diverse tasks.
Implement hybrid instant/deep reasoning to balance rapid response with thorough analysis.
Integrate Gemini 2.5 Pro for advanced multimodal reasoning (text, simulated sensor data) in data center optimization.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
The solution will be evaluated on its ability to accurately identify and propose optimizations for simulated data center scenarios (e.g., energy spikes, potential hardware failures), its effective use of Gemini 2.5 Pr...
Challenge input
{ "telemetry": { "current_power_draw_mw": float, "server_rack_temps_c": {"rack_A": float, "rack_B": float}, "utilization_percent": {"cpu": float, "...
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{ "proposed_actions": ["string"], "estimated_savings_kwh": float, "reasoning_trace": ["string"], "agent_collaboration_summary": "string" }
- Verifies if the proposed `estimated_savings_kwh` meets or exceeds a target percentage of the `current_power...
- Checks if `potential_failure` and `recommended_action` are correct based on the simulated multimodal input...
- MultimodalReasoningQuality target: 0.9
- Python execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- Verifies if the proposed `estimated_savings_kwh` meets or exceeds a target percentage of the `current_power_draw_mw` (e.g., 5% of 100.5MW -> 5025 kWh).
- Checks if `potential_failure` and `recommended_action` are correct based on the simulated multimodal input and thresholds.
Metrics
- MultimodalReasoningQuality · Target 0.9
- AdaptiveBudgetEffectiveness · Target 0.7
- RAGRelevanceAndAccuracy · Target 0.85
Workflow tasks
- Task 1
Input: { "telemetry": { "current_power_draw_mw": float, "server_rack_temps_c": {"rack_A": float, "rack_B": float}, "utilization_percent": {"cpu": float, "gpu": float}, "workload_type": "string" }, "energy_price_per_kwh": float, "optimization_goals": ["string"] }
Output: { "proposed_actions": ["string"], "estimated_savings_kwh": float, "reasoning_trace": ["string"], "agent_collaboration_summary": "string" }
Execution harness: Python
- Task 2
Input: { "telemetry_history": [{"timestamp": "datetime", "fan_rpm": int, "vibration_level": float, "power_draw": float}], "current_multimodal_input": {"image_url": "url", "audio_data": "base64_string", "logs": ["string"]}, "maintenance_thresholds": {"vibration_high": float, "fan_low": int} }
Output: { "potential_failure": "string|null", "confidence_score": float, "recommended_action": "string", "predicted_failure_date": "date|null", "agent_decisions": ["string"] }
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.