The original title is: "SafetyCommander: an AI safety officer where the model reasons and the code never decides"

Short summary

SafetyCommander is an autonomous EHS agent monitoring factory floors to detect safety hazards by reasoning over video and written policy. Core design insight: risk decisions happen in exactly one place (the VLM), while perception (YOLO), retrieval (RAG), and action modules remain decision-free and auditable. Includes practical engineering stories on detector tuning and compliance grounding via OSHA clause citations.

• •Autonomous EHS agent for factory safety monitoring powered by VLM reasoning over CCTV
• •Strict architecture: only the model decides risk; perception, routing, and reporting modules are deliberately decision-free
• •Implementation stories on forklift detector tuning and OSHA compliance grounding via retrieval

Generated with AI, which can make mistakes.