AI Alarm Analyzer Service

Express-based backend that classifies responder messages into fire alarm levels using OpenAI gpt-5-mini.

Features

• Single POST endpoint: /analyze-message
• Input JSON: { "message": "text of the responder message" }
• Returns JSON: { "alarm_level": "...", "reason": "..." }
• Auth: Uses OPENAI_API_KEY from environment (never hardcoded)
• Deployable: Ready for Vercel (serverless) or Railway

Fire Alarm Mapping (used in system prompt)

• First Alarm: 2–3 houses
• Second Alarm: 4–5 houses
• Third Alarm: 6–7 houses OR high-rise affected
• Fourth Alarm: 8–9 houses OR high-rise affected
• Fifth Alarm: 10–11 houses OR high-rise affected
• Task Force Alpha: ~12 houses
• Task Force Bravo: ~15 houses
• Task Force Charlie: fire affecting significant part of area
• Task Force Delta Echo Hotel India: fire affecting significant part of area
• General Alarm: fire affecting major part of area

Requirements

• Node.js 18+

Setup

1. Install dependencies:

   npm install

2. Create an environment file:
   • Copy env.example to .env and fill in values, or set variables in your shell.

   OPENAI_API_KEY=your_openai_key
   PORT=3000

Run locally

npm start
# or with live reload
npm run dev

Test the endpoint:

curl -X POST http://localhost:3000/analyze-message \
  -H "Content-Type: application/json" \
  -d '{"message":"Reports of 5 houses affected on Oak Street"}'

Example response:

{
  "alarm_level": "Second Alarm",
  "reason": "Responder reported 5 houses affected"
}

Deploy

Vercel

• vercel.json routes /analyze-message to api/analyze-message.js.
• Set OPENAI_API_KEY in Vercel project settings.
• Deploy via Vercel CLI or Git integration.

Railway

• The Express server (src/server.js) runs with npm start.
• Set OPENAI_API_KEY in Railway variables.

Notes

• Secrets are never hardcoded; use environment variables.
• Errors return JSON with an error field and may include details.