AetherGrid is an autonomous agentic system engineered to revolutionize renewable energy operations and maintenance (O&M) through advanced multimodal AI. Built on the Gemini 3 Pro model family, it addresses the critical challenge of reactive maintenance by enabling predictive, data-driven decision-making that reduces operational costs and maximizes energy yield.

AetherGrid leverages the full capabilities of the Gemini 3 Pro model to create a "Mission Control Orchestrator" that goes beyond simple chatbots.

• Reasoning Engine (thinking_level: high): We use Gemini 3's advanced reasoning to distinguish between environmental noise (e.g., wind gusts) and genuine component fatigue in SCADA telemetry, reducing false positives by an estimated 40%.
• Multimodal Perception: The Perception Lab module feeds 4K drone inspection footage directly to Gemini 3. The model identifies micro-cracks, leading-edge erosion, and thermal hotspots that traditional computer vision algorithms miss.
• Context Preservation: We implement a Thought Signature pattern to maintain context across multi-step maintenance workflows (Detect -> Analyze -> Schedule -> Dispatch), preventing "drift" during long-horizon tasks.
• Agentic Workflow: The system operates as an autonomous loop, using Gemini's tool-calling capabilities to query sensor data (analyze_scada_telemetry), check inventory, and dispatch repair crews (dispatch_repair_crew) without human intervention for routine issues.

AetherGrid is built on the Antigravity concept—a multi-agent framework where specialized agents collaborate to solve complex problems.

graph TD
    subgraph "Frontend (Mission Control)"
        UI[React Dashboard]
        Video[Perception Lab]
        Map[Geospatial Command]
    end

    subgraph "Agent Core (Gemini 3 Pro)"
        Orchestrator[Mission Control Orchestrator]
        Reasoning{Thinking Engine}
        Context[Context/History Manager]
    end

    subgraph "Services & Tools"
        SCADA[SCADA Analytics Service]
        Logistics[Logistics & Repair Service]
        Solar[Solar Potential Service]
    end

    subgraph "Data Sources"
        Sensors[(Telemetry DB)]
        Drones[Drone Video Feed]
        Inventory[(Parts Database)]
        Weather[Weather API]
    end

    UI --> Orchestrator
    Video --> Orchestrator
    
    Orchestrator <--> Reasoning
    Orchestrator <--> Context
    
    Orchestrator -- "analyze_scada_telemetry" --> SCADA
    Orchestrator -- "dispatch_repair_crew" --> Logistics
    Orchestrator -- "query_solar_potential" --> Solar
    
    SCADA <--> Sensors
    SCADA <--> Weather
    Logistics <--> Inventory
    Logistics <--> Weather
    Video -.-> Drones

The central intelligence hub. It continuously monitors asset health, correlates visual defects with sensor anomalies, and executes repair workflows autonomously.

• Real-time Decision Making: "Looping" agent architecture that thinks before it acts.
• Transparent Reasoning: Visualizes the AI's "Thought Process" step-by-step for operator trust.

• Visual Anomaly Detection: Upload drone footage to identify leading-edge erosion or solar panel delamination.
• 3D Digital Twin: Maps detected defects onto a 3D model of the asset for precise localization.

• Automated Dispatch: Generates work orders with estimated parts and labor hours.
• Inventory Check: Verifies part availability before scheduling.

• Node.js 18+
• A Gemini API Key (Get one at Google AI Studio)

1. Clone the repository

   git clone https://github.com/yourusername/aethergrid.git
   cd aethergrid

2. Install dependencies

   npm install

3. Configure API Key

   • Create a .env file or use the UI configuration.
   • VITE_GOOGLE_API_KEY=your_key_here

4. Run Locally

   npm run dev

1. Open the Mission Control dashboard.
2. Wait for the SCADA Simulation to trigger an anomaly (or manually inject one via the console).
3. Watch the Agent Overlay appear in the bottom right.
4. Observe the agent:
   • Think: "I see a vibration spike. I should verify this."
   • Act: Calls analyze_scada_telemetry.
   • Think: "Confirmed. This looks critical. Dispatching crew."
   • Act: Calls dispatch_repair_crew.
   • Result: A Work Order is created and displayed.

• Human-in-the-Loop: Critical repairs require manual approval.
• Simulation Mode: Runs safely without an API key using heuristic fallbacks.

MIT License. Built for the Gemini 3 Hackathon.