Medical
Symptoms, severity, nearby hospitals, and immediate guidance.
Hack Karnataka / Emergency response
Dispatch X
An AI-powered emergency dispatch system for individual calls, specialist routing, and city-scale alerts.
LIVE CALL
“There has been an accident near KLE Tech. Someone is hurt.”
- Intent
- Medical emergency
- Location
- Resolved
- Route
- MED-01
30hfrom problem to working dispatch system
The hackathon build
Emergency intake needs a useful decision before it needs another dashboard.
Our team built Dispatch X during a 30-hour hackathon around a simple problem: a caller in distress should not have to understand an IVR tree or repeat the same emergency to multiple departments.
I worked on the agent routing and backend path that turns a natural-language call into a specialist handoff. We extended the same system to city-scale voice and SMS broadcasts with live delivery and severity monitoring.
After triage
Why we split one voice agent into four jobs.
Police
Threat context, location, escalation state, and a bounded handoff.
Fire
Incident type, nearby response services, and dispatch information.
30-hour notes
What we chose under hackathon pressure.
The architecture was shaped by a working deadline. These choices got a complete call through the system, but they also left clear follow-up work.
Route before asking every question
The first agent only identifies the emergency class and preserves context. Domain-specific prompts and tools live with the specialist agent that needs them.
Persist the raw call and the analysis
Gemini can add severity and classification, but the transcript, provider events, cost, and tool results remain available for inspection.
Keep escalation visible
WebSockets make the call observable in the dashboard, while an operator can still intervene instead of trusting an agent to close the loop alone.
Beyond one call
100concurrent broadcast calls with live operational state.
We added concurrent voice alerts, SMS follow-ups, and a live view of transcripts, delivery, severity, and escalation state.
My part of the team build
AI systems and backend engineer
I built across the agent routing, FastAPI services, emergency integrations, live call state, and the workflows connecting voice conversations to actionable dispatch information.
- Our team started with one incoming call and built outward. I worked on the VAPI routing path, the FastAPI services behind tool calls, and the state that connects a transcript to a dispatch action.
- I wired specialist medical, police, and fire flows to nearby-service lookup, call transfer, SMS follow-up, and the MongoDB call record.
- I also worked on the live operational side: WebSocket updates, transcript and status handling, and the batch-broadcast path used for city-scale alerts in the demo.
FastAPIVAPIMongoDBTwilioGeminiWebSockets