Agentic AI for Predictive Maintenance using OpenVINO

This project aims to develop an agentic AI system for predictive maintenance of urban infrastructure using OpenVINO. The goal is to enable intelligent monitoring and decision-making by combining multimodal perception with reasoning capabilities on edge devices. The proposed solution integrates a vision-language model to detect defects (Deformation, Obstacle, Rupture, Disconnect, Misalignment, Deposition) from visual data and a language model to interpret these defects in terms of severity, risk, and maintenance priority. These components will be connected through an agentic pipeline that performs perception, reasoning, and action in a unified workflow. The system will be implemented using OpenVINO-optimized models and deployed on Intel edge hardware to ensure efficient, low-latency inference. Open datasets will be used to validate the system in realistic scenarios. Expected deliverables include: - An end-to-end agentic AI pipeline for predictive maintenance - Integration of multimodal perception and LLM-based reasoning modules - Deployment and optimization using OpenVINO on edge platforms - Benchmark results evaluating performance and system effectiveness - Comprehensive documentation and reproducible codebase