ExitMatrix.
A fire evacuation system for buildings. Staff draw floor plans on the web, an AI model plans escape routes around the fire, and a mobile app guides people out with live voice directions.
Lead author & full-stack engineer
- Mobile (Flutter & iOS)
- AI & Automation
- Flutter
- Unity (C#)
- Python (NumPy, FastAPI)
- Node.js (Express)
- Firebase Firestore
- Firebase Auth
- Firebase Hosting
- Q-learning (RL)
Fixed exit signs cannot react when a fire blocks a corridor. I built ExitMatrix, a research system that plans a new escape route the moment the fire map changes. A web editor stores each floor as a grid in Firebase Firestore, a Python service runs a trained Q-learning agent, a Node.js engine adds fire safety buffers and turn-by-turn steps, and a Flutter app speaks each step aloud. I used Firestore realtime listeners as the shared link so every part reacts to the same live map with no polling. A Unity 3D building let me test routes safely.
- 01Learns the way out. I trained a Q-learning agent for 100,000 episodes that rewards reaching an exit and penalises fire, walls and nearby danger cells.
- 02Keeps people away from flames. An A* engine adds a safety buffer around every fire, crosses it only when no other route exists, and warns the user if it must.
- 03Directions people can follow. Each route becomes plain steps like 'Go straight 40 meters and turn left', read out by text-to-speech and sent as alerts.
- 04One live map, no polling. The web editor, ML service, Node.js backend and Flutter app all react to the same Firestore document in real time.