Inspiration 💡
Industrial sites and construction zones often lack 24/7 safety oversight, leading to preventable injuries. We were inspired to build a tool that acts as a "second pair of eyes" for HSE (Health, Safety, and Environment) officers, using vision AI to ensure every worker returns home safely.
What it does ✨
SafeStep AI is a specialized HSE auditing dashboard that analyzes industrial environments in real-time:
🛡️ PPE DETECTION: Scans for helmets, vests, and safety gear with high accuracy.
⚠️ HAZARD IDENTIFICATION: Flags critical violations, such as workers positioned beneath suspended loads.
📊 SAFETY SCORING: Generates an immediate HSE Audit Report with numerical safety scores (e.g., 40/100 for high-risk zones).
🛠️ MAINTENANCE 4.0: Provides actionable recommendations to mitigate technical risks and improve workflow safety.
How we built it 🏗️
Vision Engine: Powered by Gemini 1.5 Flash for high-speed image and video analysis.
Frontend: Built with React to create a responsive, dark-mode auditing interface.
Development: Hosted and refined on Replit, utilizing its AI-integrated workspace for rapid debugging.
Challenges we ran into 🚧
During development, we encountered a significant runtime error with the react-markdown component because it no longer supported the className prop directly. We had to debug the component structure and wrap the content in a standard div to maintain our custom styling while ensuring the app hot-reloaded cleanly.
Accomplishments that we're proud of 🏆
Successfully grounding the AI to provide specific justifications for safety scores, such as identifying "suspended excavator buckets" as imminent hazards.
Creating a "Verified Status" workflow that allows for human-in-the-loop oversight.
What we learned 🧠
We learned the importance of prompt engineering for vision models—specifically how to guide the AI to focus on "Maintenance 4.0" parameters rather than just general object detection. We also mastered troubleshooting React component versioning in a live production environment.
What's next for SafeStep AI 🚀
Live Stream Integration: Moving from static media uploads to direct CCTV feed monitoring.
Predictive Risk Modeling: Using historical audit data to predict when a site is most likely to experience a safety breach.
Log in or sign up for Devpost to join the conversation.