ConstructGuard AI

With ConstructGuard AI, we’re not just detecting safety gear — we’re protecting lives, empowering workers, and building a safer future powered by intelligence.

Comment

🦺 ConstructGuard AI – Agentic PPE Compliance & Risk Monitoring System

💡 Inspiration

Although construction workers make up just 6% of the US labor force, they account for nearly 20% of all workplace fatalities, with falls, slips, and trips making up over 38% of these deaths—costing the industry over $11.5 billion each year in injury and fatality-related expenses. That insight led us to build an AI-powered safety monitoring system that transforms ordinary CCTV footage into a real-time safety intelligence platform — one that detects non-compliance instantly, alerts supervisors on site, and provides insurers with live compliance metrics per location. Our motivation was simple: If we can detect danger before it happens, we can save lives — and make construction safety proactive, not reactive.

⚙️ What It Does

  • Detects PPE compliance (helmet, vest, gloves) in real time using YOLO.
  • Sends instant alerts for violations.
  • Calculates a dynamic risk score based on safety behavior.
  • Provides data for adaptive insurance pricing and compliance analytics.

🛠 How We Built It

  • Trained a custom YOLOv11n model on PPE datasets.
  • Used OpenCV for frame extraction and annotation pipeline.
  • Developed a compliance engine that classifies risk levels per frame.
  • Integrated an Agentic AI module to simulate autonomous decisions and alerts.
  • Output generated as an annotated video + risk report summary.

⚡ Challenges We Ran Into

  • Managing large video data and maintaining inference speed.
  • Balancing accuracy vs latency for real-time detection.
  • Fine-tuning confidence thresholds to reduce false alerts.
  • Conceptualizing a dynamic insurance model linked to compliance metrics.

🏆 Accomplishments We’re Proud Of

  • Achieved >93% detection accuracy with near real-time performance.
  • Designed a scalable AI architecture that supports agentic decision loops.
  • Demonstrated dynamic premium adaptation using risk scores.
  • Built a production-ready pipeline that could run on edge devices.

📚 What We Learned

  • How Agentic AI can transform traditional detection into autonomous decision systems.
  • The importance of data-driven safety insights for business and insurance.
  • Hands-on mastery of YOLO optimization, OpenCV streaming, and automation pipelines.

🚀 What’s Next for ConstructGuard AI

  • Integrate with IoT sensors for multi-modal risk detection.
  • Build a real-time web dashboard for compliance visualization.
  • Implement LLM-driven report generation for insurance documentation.
  • Deploy on edge devices (Jetson Nano) for live industrial sites.
  • Expand into predictive safety analytics using reinforcement learning.

Built With

Share this project:

Updates