Inspiration Industrial environments—from construction sites to manufacturing plants—are inherently dangerous. Despite CCTV coverage, human supervisors cannot monitor every feed 24/7. We were inspired to build Guardian Sight AI to bridge this gap, moving from passive recording to active, "thinking" safety auditing that can prevent accidents before they happen.
What it does Guardian Sight AI is an autonomous safety agent. It monitors live video feeds or image uploads and uses Gemini 3 to:
Identify PPE Violations: Checks for helmets, vests, and gloves.
Assess Risk Situations: Recognizes dangerous behaviors, such as a worker standing too close to a ledge or operating machinery without a spotter.
Reason Through Safety Protocols: It doesn't just "see" objects; it understands context (e.g., "The worker is 15ft up and the harness is not tethered").
Trigger Instant Alerts: Uses Function Calling to notify supervisors immediately when a life-threatening risk is detected.
How we built it Frontend: Built with React (leveraging AI Studio's "Build" mode) for a clean, professional supervisor dashboard.
Backend: Node.js/Python server handling the video stream and API orchestration.
AI Core: Gemini 3 Flash for low-latency real-time video processing.
Logic: We used Thinking Level: High to ensure the model performs multi-step reasoning before triggering an alert.
Challenges we ran into The biggest challenge was reducing latency. Real-time safety needs fast responses. We optimized this by using the Flash model and implementing a "Sampling" logic to send the most relevant video frames for high-context analysis.
Accomplishments that we're proud of Successfully implementing Multimodal Function Calling. Seeing the AI detect a missing harness in a video and automatically trigger a "Safety Alert" log on the dashboard was a major milestone.
What we learned We learned how to prompt Gemini 3 to act as an "Agent" rather than just a chatbot. The difference is the "Thinking" mode, which allows the model to follow complex industrial safety regulations (OSHA) accurately.
What's next for Guardian Sight AI We plan to integrate with wearable IoT devices to trigger haptic feedback (vibrations) on a worker's smartwatch the moment the AI detects they are in a high-risk zone.

Log in or sign up for Devpost to join the conversation.