Project information
- Category: AI Agents, Sustainability, Azure Cloud, LLM Integration, Applied AI
- Project date: April 2025
- GitHub: SentinelGreen
- Collaborators: Neha Pawar
Introduction
SentinelGreen is a multi-agent, LLM-integrated AI system designed to optimize energy consumption, cooling, compliance, and cybersecurity within modern data centers. It addresses critical challenges in data center management through autonomous decision-making and real-time analytics.
- Autonomous Agent Collaboration
- Energy Optimization & Sustainability
- Cybersecurity Threat Detection
- Predictive Maintenance
- Compliance Monitoring
- Resource Allocation Optimization
By leveraging Microsoft Azure services and lightweight language models like Phi-3 Mini, SentinelGreen combines real-time sensor telemetry with natural language reasoning to deliver proactive insights and autonomous decision-making.
The system provides full flexibility—agents can operate locally for testing via FastAPI and Chainlit, or scale to production using Azure's IoT, AI, and analytics ecosystem. Built with a modular architecture, SentinelGreen serves as a blueprint for future-ready, sustainable, and secure digital infrastructure.
Objective
The primary objective of SentinelGreen is to transform how modern data centers manage energy resources, respond to operational risks, and achieve environmental sustainability goals. Data centers are among the most energy-intensive infrastructures globally, contributing significantly to electricity consumption and carbon emissions, while simultaneously being vulnerable to cybersecurity threats, hardware failures, and inefficiencies.
SentinelGreen addresses these challenges by creating an intelligent, multi-agent system that works collaboratively to continuously monitor and optimize data center operations. The system aims to reduce energy consumption, enhance security posture, prevent equipment failures through predictive maintenance, ensure regulatory compliance, and optimize resource allocation—all while providing transparent decision-making processes that can be audited and verified.
Process
The development and operation of SentinelGreen involves several interconnected processes:
- Agent Architecture Design: Created six specialized autonomous agents with distinct but complementary roles.
- Prompt Engineering: Developed tailored prompts for each agent to guide LLM decision-making processes.
- System Message Definition: Established clear system messages that define each agent's operational boundaries.
- Telemetry System Integration: Built IoT connectivity for real-time sensor data ingestion from data centers.
- Azure Service Configuration: Set up IoT Hub, Stream Analytics, and Data Lake storage for data pipeline.
- LLM Integration: Implemented Phi-3 Mini model deployment for lightweight, efficient reasoning.
- Local Testing Framework: Created FastAPI-based local LLM service for offline development and testing.
- Data Simulation: Developed device simulators to generate realistic telemetry for test environments.
- Stream Processing: Configured Stream Analytics jobs to transform raw telemetry into actionable inputs.
- Agent Communication Layer: Built inter-agent communication protocols for collaborative decision-making.
- Decision Logging System: Implemented comprehensive logging for audit and improvement purposes.
- UI Development: Created Chainlit-based interactive dashboard for agent testing and visualization.
- CLI Interface: Built command-line interface for rapid agent testing and development.
- Pipeline Testing: Verified end-to-end data flow from sensors to decisions and back to systems.
- Documentation: Created detailed documentation of system architecture, agent behaviors, and integration points.
- Edge Deployment Strategy: Designed approach for running agents close to data sources for lower latency.
Tools and Technologies
SentinelGreen leverages a comprehensive technology stack to enable its autonomous agent operations:
Microsoft Azure Services:
- Azure IoT Hub
- Azure Stream Analytics
- Azure Synapse Analytics
- Azure Data Lake Gen2 Storage
- Azure AI Studio (AI Foundry)
LLMs and Agents:
- Phi-3 Mini
- AutoGen Framework
- ConversableAgent Implementation
Development Tools:
- Python
- FastAPI
- Chainlit
- JSON
Testing and Simulation:
- Device Simulator
- Local LLM Service
- Mock Data Generator
Riding the AI Agent Wave
SentinelGreen provides substantial value across multiple dimensions of data center operations:
The system reduces operational costs by optimizing energy consumption through real-time monitoring and adaptive control. By implementing predictive maintenance, it prevents costly downtime and equipment failures before they occur. The integrated security monitoring enhances protection against physical and cyber threats through continuous vigilance and anomaly detection.
From a sustainability perspective, SentinelGreen directly addresses carbon footprint reduction through efficient resource utilization and intelligent cooling management. The compliance auditing features ensure adherence to evolving environmental regulations and sustainability commitments, while providing transparent reporting for stakeholders.
The modular architecture allows for flexible deployment options—from small test environments to enterprise-scale data centers—with consistent decision logic across scales. By combining lightweight language models with structured data analysis, SentinelGreen demonstrates how AI agents can transform infrastructure management while maintaining human oversight and explainable decision processes.