AI-powered emergency call center for Berlin that handles live 112 calls through conversational voice AI.
The dispatcher conducts natural emergency interviews, triages incidents by severity, provides real-time first-aid guidance via RAG-powered protocol search, and autonomously dispatches ambulance/police/fire units with live ETA tracking and animated map visualization.
| Component | Technology |
|---|---|
| Voice Pipeline | LiveKit Agents + Silero VAD |
| STT | Groq Whisper Large v3 Turbo |
| LLM | Kimi K2 Instruct (128k context) |
| TTS | ElevenLabs Turbo v2.5 |
| RAG | Redis Vector DB + OpenAI Embeddings |
| Frontend | Leaflet.js + OpenStreetMap + WebRTC |
8 functions chained seamlessly during live conversation:
- Address geocoding
- Unit dispatch (ambulance/police/fire)
- Incident severity assessment
- Caller GPS location
- Emergency protocol search
- ETA calculation
Template-driven prompts enable rapid iteration. The AI provides CPR coaching while dispatching help, turning passive call-taking into active life-saving guidance.
- Reduces dispatcher workload
- Ensures protocol consistency
- Enables 24/7 coverage
- Cuts response coordination time from 30-90s to under 2 seconds