Firefly

A survey conducted by Nasdaq reveals a stark reality: 67% of women feel unsafe walking alone at night, 70% wish for a discreet safety device, and 50% want a simple way to send their live location to close contacts. Driven by personal resonance with these statistics, our team came together to build Firefly, a web app designed to act as an attentive virtual companion on your walk home. Firefly keeps you company by chatting in a warm, comforting voice while displaying your route on an interactive map. Crucially, it continuously monitors your safety in the background for a scream or a custom secret code word; if triggered, it initiates a silent countdown that automatically sends your live location to trusted contacts unless manually canceled.

We built Firefly using Google Gemini for conversational intelligence and real-time speech transcription (Gemini Live), which catches the secret code phrase; ElevenLabs for voice synthesis; a scream-detection model we trained in Python (TensorFlow, scikit-learn and the YAMNet audio model) that runs directly on the phone with TensorFlow.js; Tiger Data (PostgreSQL + TimescaleDB) for structured and time-series data storage, hosted on Azure (East US 2 (Virginia)); AWS Lambda to host the app and API; Vonage for the emergency call and Twilio for text alerts; React, Vite and Tailwind CSS for the front end and Node.js for the back end; and Leaflet with OpenStreetMap for maps, walking directions and place search. We wrote it all in Microsoft VS Code with Google Antigravity.

We are incredibly proud that our diverse group completed every individual component and merged them into an impactful, functional project. Through this process, we gained hands-on experience integrating multiple APIs and applications simultaneously under tight deadlines. Looking ahead, we plan to expand Firefly into an everyday virtual audio journal that uses Google Gemini Agents to process user requests, Eleven Labs to generate real-time audio responses, and Tiger Data to remember past user conversations to ask thoughtful follow-up questions, creating a truly dual-purpose safety and wellness platform.