Guardian
VozGuard AI uses real-time transcription and OpenAI to triage emergency calls, empowering operators with intelligent risk assessments and location tracking.
Project Description
VozGuard / SilentGuard AI is an intelligent emergency call triage system that receives real-time transcriptions from a conversational voice agent (voice bot) and automatically classifies the content to support human operators in decision-making.
The flow works as follows:
A conversational agent answers phone calls and transcribes the audio in real time.
At the end of (or during) the call, a webhook sends the full payload — transcript, call metadata, caller geolocation, sentiment analysis, classification, and routing — to the /api/webhook/call-transcript endpoint.
The system normalizes the payload, classifies the emergency via AI (domestic violence, medical emergency, fall/accident, missing person, robbery/assault, fire, possible hoax, out of scope), estimates the geographic location from the text, and generates a structured occurrence with risk level, reliability score, detected signals, and recommendations.
The operational cockpit displays in real time: formatted transcript, classification, map with estimated location, nearby support points (hospitals, police stations), immediate risk checklist, event timeline, and suggested action.
An AI copilot (CopilotKit) assists the operator with classification explanations, operational briefings, and actions such as requesting human validation or marking as out of scope.
The system is designed for scenarios where the victim cannot speak freely (silent mode), detecting discreet codes and indirect distress signals. The final decision is always human — AI supports but never dispatches real emergency services.
Stack: Next.js 14, React 18, CopilotKit, SQLite (local persistence), OpenAI (classification), deployed via Docker on EasyPanel.
Prior Work
Our team built this project during the hackathon. We did not build upon a previously completed product, proprietary codebase, or production-ready system.
During the hackathon, we created the product concept, solution design, CopilotKit-based interface, frontend screens, backend/webhook integration flow, data structure, documentation, and presentation materials.
The project was built using existing open-source frameworks, libraries, and development tools, including CopilotKit and related frontend/backend technologies. However, the core solution, user experience, integration logic, and final prototype were created specifically for this hackathon.