Project Story
Inspiration
Construction workers die at 5x the rate of other industries, yet they avoid reporting safety incidents due to complex paperwork and time constraints. In Sydney, 35 people get seriously injured every day. I realized that voice conversation - the most natural form of human communication - could reduce this barrier. Rather than forcing workers to navigate forms on their phones while wearing gloves in harsh conditions, what if they could simply talk to an AI safety coordinator?
What I Learned
Building with ElevenLabs Conversational AI taught me the power of true voice-to-voice interaction. This isn't just text-to-speech - it's actual conversation with natural back-and-forth dialog, powered by Google Gemini 2.5 Flash for sophisticated understanding. The combination creates an experience that feels like talking to a human safety manager, making technology disappear and focusing on the actual problem: getting critical safety information reported quickly.
How I Built It
The architecture prioritizes simplicity over complexity:
- ElevenLabs Conversational Agent: Created a specialized AI safety coordinator with professional voice persona
- Google Gemini 2.5 Flash Integration: Embedded through ElevenLabs for intelligent conversation logic and automatic incident classification
- Mobile-First Web Interface: Single HTML page with embedded ElevenLabs widget - no app stores, no downloads
- GitHub Pages Hosting: Free, reliable hosting with HTTPS for microphone access
- Webhook Notifications: Safety Manager alert system for urgent incidents
The entire system runs without infrastructure - just a browser and voice conversation.
Challenges I Faced
Voice UX Design: Creating natural conversation flows that feel professional yet approachable for construction workers required multiple iterations. The AI needed to ask the right follow-up questions without feeling robotic.
Cross-Platform Compatibility: Ensuring microphone access works across iOS Safari, Android Chrome, and desktop browsers required careful HTTPS configuration and responsive design.
Cost Management: Building a working demo that judges can test while preventing runaway ElevenLabs usage costs during the hackathon required careful webhook monitoring and time-limited access.
Real-World Terminology: Training the conversation to understand construction-specific language - "scaffolding joints," "crane swing radius," "fall protection" - while maintaining natural dialog flow.
Log in or sign up for Devpost to join the conversation.