Agentic Search Engine for Real-Time Internet of Things Data.

Area of Science:

• Internet of Things (IoT)
• Artificial Intelligence
• Information Retrieval

Background:

• Fragmentation and data heterogeneity in IoT systems hinder seamless data sharing and management.
• Existing IoT search engines primarily function as device discovery tools, lacking access to application data.
• Previous efforts like IoTCrawler faced challenges due to IoT system fragmentation and data diversity.

Purpose of the Study:

• To introduce the IoT Agentic Search Engine (IoTASE), a real-time semantic search engine for IoT environments.
• To address the challenges of navigating and searching vast, heterogeneous streams of real-time IoT data.
• To enable interoperable content and sensor data sharing among collaborative IoT systems.

Main Methods:

• Leveraging Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) techniques.
• Developing a unified framework inspired by the World Wide Web for accessible information spaces.
• Implementing a hypothetical deployment in Toronto with 500 services and over 37,000 data entries for evaluation.

Main Results:

• IoTASE achieved 92% accuracy in retrieving intent-aligned services.
• The system consistently generated concise, relevant, and preference-aware responses.
• IoTASE outperformed generalized outputs from systems like Gemini in accuracy and relevance.

Conclusions:

• IoTASE demonstrates significant potential for making real-time IoT data accessible and actionable.
• The semantic search engine facilitates intelligent decision-making across diverse IoT application domains.
• The RAG and LLM-based approach effectively tackles the complexities of IoT data search.