Alarm management & dispatch · 2022 – present

AlarmHub

Alarm management, dispatch and situational awareness for public safety organisations — four apps, one backend, one shared operational picture.

Back to all projects
Role
Main developer
Timeframe
2022 – present
Context
In the AlarmHub product team
Technologies
Vue 3PrimeVueOpenLayersNode.jsExpressGraphQLRESTSocket.IOSequelizePostgreSQLS3 / MinIODockerPM2SentryBitbucket Pipelines

AlarmHub receives alarms from different sources, guides them through dispatch and makes the progress visible to everyone involved. Over four years I helped build the platform as a main developer — with 2'055 commits (all refs). I built the situation map and the mobile field app from scratch.

01

Context

Fire services and public organisations receive alarms by email, SMS, direct integrations and other systems. These sources must feed reliably into one shared process.

Beyond administration, dispatch, the situation map and field responders each need their own interface — but the same data.

02

My contribution

  • Built the live situation map (eagle, later map) and the mobile field app (mobile, later storm) from scratch and led their development.
  • Built the Storm domain for incidents, assignments, groups, resources and connections; 98 commits in the central incident controller (master; 106 across all refs).
  • Built the multi-level escalation engine with rule filters, timeouts and lost-timer recovery, and extracted the shared rule logic into its own package.
  • Helped shape alarm ingestion across 30 protocol profiles, including eAlarm variants, Intergraph, MoKoS, NovaAlert, email and SMS.
  • Co-developed GraphQL and REST, real-time updates over Socket.IO, and role and tenant checks across all four interfaces.
  • Co-designed cross-product interfaces: Intervento can consume AlarmHub data; the FloriPlan app gained alarm and app-update capabilities.
  • Isolated the libpostal address service into its own container because the 2 GB model overwhelmed the Node cluster workers.
  • Took on code reviews, architecture work and technical documentation within the team.
03

Architecture & delivery

  • Monorepo with four Vue 3 SPAs — client/, dispatch/, map/ and storm/ — plus a shared Node/Express backend.
  • GraphQL for administration and dispatch, REST for the field app, Socket.IO for real time; OpenLayers for the situation map.
  • 91 Sequelize models and 250 migrations on PostgreSQL (master; 255 on the feature branch); S3-compatible object storage for PDFs, protocols and attachments.
  • Socket.IO cluster adapter with sticky sessions and rooms per app, organisation and user.
  • Separate containers for the application and the libpostal sidecar, PM2 cluster, Sentry and Bitbucket Pipelines.
System design From source to situation
Alarm ingestion
Interfaces
Data
Alarm sources email · SMS · direct
30 protocol profiles normalise · map
Inbound pipeline validate · dedupe
Dispatch & escalation rules · timers
API layer GraphQL · REST
Socket.IO cluster · sticky
Four interfaces client · dispatch · map · storm
libpostal address sidecar
PostgreSQL 91 models
MinIO / S3 files
30 profiles Normalised GraphQL / REST Live Files Address lookup

Alarms arrive through direct integrations, email and SMS. 30 protocol profiles normalise the inputs before dispatch, escalation and the four interfaces use the same state.

Design decisions
  • 30 protocol profiles bundle heterogeneous alarm sources behind a normalised inbound layer.
  • GraphQL for administration and dispatch, REST for the field app, Socket.IO for live updates.
  • libpostal runs as its own container because the 2 GB model overwhelmed the Node cluster workers.

Scroll horizontally to see the full design.

04

The hard parts

01

Real time across a cluster

Socket.IO breaks once several PM2 workers are running. The cluster adapter with sticky sessions and rooms per app, organisation and user keeps dispatch, map and field app on the same state.

02

Alarms from many worlds

30 protocol profiles for cantonal eAlarm variants, Intergraph, MoKoS, NovaAlert, email and SMS normalise different formats. Address lookup with libpostal needed 2 GB and had to move out of the Node cluster into its own container.

03

Escalation that survives restarts

Multi-level escalation needs reliable timeouts. State lives in the database, in-memory timers speed up the normal case, and a cron job recovers timers lost after a restart.

05

Outcome

  • One shared operational picture for control room, dispatch and field responders.
  • Alarm, dispatch, field assignment and report in one continuous flow.
  • Escalations run without manual follow-up; lost timers are recovered automatically.
  • One integration layer for 30 protocol profiles and several output channels.

Contact

Let's talk about the next role.

I am looking for a position where I can take responsibility for systems and teams as a lead engineer or software architect. Ideally where reliability, data and real-world workflows meet.

Send an email [email protected]

Availability

Region
Thun/Bern
Workload
80%
Work location
On-site, hybrid or remote by arrangement
Travel
Generally yes, scope depending on the role and responsibilities
Start
by agreement