1 of 6

RouteZero

(AI Dispatch Router)

Ved Patel and Vedant Daga

MakeNJIT26

2 of 6

Problem Statement

The main problem is that during natural disasters, emergency dispatch breaks down. First, responders across multiple agencies operate on manual radio communication and human coordinators to route help across a disaster zone spanning dozens of square miles. With hundreds of simultaneous emergencies, this system collapses.
Our target audience are first responders, volunteers, and emergency management agencies operating in natural disaster scenarios. Ultimately, the people most affected are the civilians in the disaster zone whose survival depends on how fast and accurately help is routed to them. Our solution is designed to keep costs as low as possible while still providing meaningful assistance, ensuring that every firehouse, police station, EMS responder, and emergency team has the resources they need to save lives.
The main goal of the project is to help first responders with Natural Disaster Response
This is important because coordination failure during natural disasters is documented and deadly. Hurricane Katrina knocked out 38,911 call centers and left over 2,000 responders sharing just 3 radio frequencies. As disasters grow in frequency and scale, the gap between what traditional dispatch can handle and what disasters actually demand will only widen. RouteZero addresses that gap.

Problem:

During natural disasters, communication systems break down

Radio and Human Dispatcher are both prone to breaking down and can make many mistakes under high pressure situations

Affects large areas at a time, often hundreds of sq mi

Coordination between different agencies suffers when coordinating over large areas

Along with the stress on dispatches, mistakes, time wasted, and failed coordination all lead to injuries and sometimes fatalities
Our target audience are first responders, volunteers, and emergency management agencies operating in natural disaster scenarios. Ultimately, the people most affected are the civilians in the disaster zone whose survival depends on how fast and accurately help is routed to them. Our solution is designed to keep costs as low as possible, ensuring that every firehouse, police station, EMS responder, and emergency team has the resources they need to save lives. <- we did this by making every sensor modular, using common objects found at home like bluetooth headphones or webcams.

Main Goal

to help first responders during a natural disaster by assiting them in their response

Important because

We keep costs down because we use a modular design. u can use any webcam, any earbuds, and can plug and play the sensors.

3 of 6

Project Description

In summary, RouteZero is an AI-powered dispatch system built on Raspberry Pi field units deployed across a disaster zone. A responder speaks a request, it gets transcribed, reasoned over by a central LLM, and the right responders are informed instantly. Every 5 minutes, each unit sends a heartbeat so the AI always has a live picture of the entire operation. Long-term memory ensures it never loses context over a hours-long response.
Our solution is unique because most existing dispatch tools are built for single-incidents. RouteZero is built specifically for the chaos of natural disasters: multi-agency, multi-incident, large geographic scale. First, it requires no human coordinator. Second, it scales like a pyramid: one LLM manages a first responder, another manages multiple of those llms, another manages multiple regions. The same architecture works for a local wildfire or a national disaster. Third, it remembers: unlike a standard LLM that often forgets after each call, RouteZero maintains situational awareness across the entire duration of a response, so it never contradicts an earlier decision or loses track of where teams have already been deployed. On the hardware side, every sensor is modular so that if one somehow happens thru the protective barriers, they can be easily replaced and tested with ease, saving time and money..

In summary:

It’s an AI-powered dispatch system built on Raspberry Pi field units
These units are spread across the disaster zone
Walk through the flow:
A responder speaks a request
It gets transcribed into text
Then analyzed by a central AI (LLM)
And instantly routed to the right responders while providing them a real time update of conditions to expect
Highlight real-time awareness:
Every unit sends a heartbeat every 5 minutes
So the system always knows what’s active and where
This gives a live, constantly updating view of the entire operation
The system has long-term memory
It keeps context over hours-long responses
So nothing gets lost or repeated

Unique Because:

The system chooses the “right” responders using the LLM
It doesn’t make random assignments—it’s context-aware and data-driven
It uses previous memory like past injuries, team availability, and environmental conditions
It also pulls in real-time sensor data from all field units
This combination allows it to understand both history and the current situation before routing help
The same architecture works whether it’s a local wildfire or a national disaster
The only difference is scale
You simply add more layers to the pyramid as the operation grows
This makes the system flexible enough to handle anything from small incidents to large-scale disasters and can be updated in seconds by replacing the sensors using the modular design

4 of 6

Technical Description

Some items we brought from home are the:

Raspberry Pi 5: the main part to the field unit
AirPods Pro: audio input/output, very sustainable
Neo6M (GPS Module): another sensor, for location
ADS1115: an analog to digital converter for pi

The software we used in our project are:

Python: the main programming language
FastAPI: backend framework for Python
Typescript + React: Language + framework for the frontend
Grok: used for fast llm inference
OpenAI Whisper: on-pi speech transcription

The biggest challenge was getting reliable Bluetooth audio on Raspberry Pi. ALSA configuration errors, PipeWire compatibility issues, and unstable AirPods connections caused audio to drop or fail to route correctly. We resolved this by switching fully to PipeWire, setting explicit default sinks and sources via pactl, and bypassing PyAudio's device selection entirely by defaulting to the system audio device.

5 of 6

Demo

https://vimeo.com/1182388957?share=copy&fl=sv&fe=ci

6 of 6

Questions

Panelists will ask questions