First Responder Drone Dispatcher with AI

First Responder Drones with AI

Inspiration

We are building an "First Responder Drone Dispatcher with AI" - emergency response system. We were inspired by the opening day talk by Justine Johnson on how Michigan and Detroit are thinking about using AI and Drones for citizens. Next day we met Capt Sculley at the Firestation right across TechTown to understand how the Fire services would benefit from a system, and he gave us many scenarios to think about. The Goals of this system are:

Rapid situational awareness as the teams are getting dispatched
Smart dispatch: Orchestrate the right amount of response from EMS/Fire/Police

What it does

The AI edge service on the Drone analyzes images from a drone to detect incidents (fire, accidents, casualties, roadblocks) and emits a structured scene report. The backend service uses an LLM/CrewAI orchestrator to determine which units to dispatch (EMS, Fire, Police) and can generate clarifying questions per agent to improve decisions. The dispatcher console looks at all the incoming calls, and the images captured by the drone, the real time analysis, and the scene report, and calls the AI agents for various services. These agents, can then give the teams situational awareness.

How we built it

We have build the two services (Edge and Backend) in CursorAI. The front end UI is build using Lovable and Supabase. We are using 4 AI Models for various use cases, and we tested with 15-20 different scenarios. The backend system is fully secure with OAuth, and the front end is secure with RLS in supabase. All the code is in github repo, the two services are deployed Hostinger and the UI is deployed in Lovable.

Challenges we ran into

We ran into multiple challenges from finding the right models, to the size of the models, hardware considerations of loading the model, package dependencies, and issues with communication between various services. We overcame all these challenges in multiple rapid iterations using Lovable, Cursor and ChatGPT.

Accomplishments that we're proud of

We are very proud of a fully working system that we built within a 1.5 days. This is fairly complex system, where we tested multiple use AI models, deployed the services and have a fully functional UI. We are also proud that we first spoke to the Captain of the fire station to get actual feedback before we wrote the first line of code.

What we learned

Many lessons, we can iterate faster with AI coding tools, but we still have to think about testing, validation, deployment, security. We broke the project in multiple smaller milestones, and tackled them as a team.

What's next for First Responder Drone Dispatcher

We want to talk to more "customers" - first responders. We want to demo this to Capt Sculley and his team, other departments, and get more feedback and discover what we don't know. One of the use cases that we discovered was "water accidents" in lakes around us. Once we have good set of use cases, we want to go for funding to built the v1 of this project.