Show HN: Designing a factory-safety agent (model reasons, code routes) (github.com)

A new AI-powered factory safety agent named SafetyCommander was unveiled during the Zapdos Labs × Antler hackathon, aimed at improving workplace safety through intelligent monitoring and decision-making. This agent autonomously oversees production floors using camera feeds and safety policy documents, assessing risk levels based on real-time data and regulatory citations, and alerting the appropriate personnel according to defined safety protocols. Significantly, it can adapt its assessments and actions just by editing the safety policy text, demonstrating a level of reasoning that distinguishes it from traditional hardcoded systems. SafetyCommander utilizes a sophisticated tech stack, including Qwen3-VL for vision-language modeling, YOLO for object detection, and a Flask web app for user interaction. The system operates within a deterministic loop—sense, think, act, and report—ensuring all decisions about risk originate from a single module. This focused architecture provides grounded and defensible safety oversight, effectively tracking workplace hazards over time, while allowing for a human-in-the-loop approach to confirm safety actions. By emphasizing temporal risk assessments rather than static evaluations, SafetyCommander represents a forward-thinking approach to automated safety in industrial environments, which could set new standards in workplace safety management.