Summary of the Pitch Frontline workers frequently face injuries and near-misses on job sites, but safety issues go unreported due to fear of retaliation, tedious paperwork, and language or literacy barriers. By providing a simple, voice-first reporting tool, workers can instantly speak about hazards—like a near-miss with a falling beam—without touching a multi-page form. Behind the scenes, AI turns that spoken input into structured, actionable data so no detail is lost. Complete anonymity eliminates the fear of getting fired, empowering workers to speak up and forcing management to take proactive action—especially when multiple anonymous reports highlight the exact same threat.
What Inspired Us Seeing the huge gap between real hazards on the ground and what actually gets reported to management. Workers on busy job sites have their hands full, deal with language barriers, and above all, fear losing their livelihoods just for speaking up about dangerous conditions. The realization that a simple, accessible voice interface paired with true anonymity could break down those barriers and literally save lives was the driving force behind this solution.
How We Built It + Challenges We started with the idea of building a voice-based safety incident tool, but pivoted after realizing that space was already crowded with existing enterprise software. Instead, we built a voice-first hazard reporting tool that focuses on surfacing risks before they become cause for actual injuries. Workers speak a hazard out loud using the blowers built-in speech recognition, which an LLM (Grok, via xAI's API) structures into category, location, and severity data stored in a database. The system then flags repeated hazards in the same area, surfacing emerging patterns on a live dashboard instead of leaving each report unattended. Our biggest challenges were technical integration issues, given that we had to pivot our entire backend to a new LLM provider mid-hackathon after losing access to our original credits, as well as debugging inconsistent JSON output from the model, and fixing browser-side speech recognition issues that were breaking our pipeline.
Log in or sign up for Devpost to join the conversation.