Part 5 of 11
Platform and data pipeline
The backend that receives every position report, keeps it safe, runs the rules and AI, and serves the staff app, the control room dashboard and the alerts.
Architecture
Small, separate services connected by an ordered event stream. Any one of them can be restarted or scaled without losing data, because fixes are stored before they are processed.
Core components
Device management
Provisions each tracker with its own certificate, binds it to a fleet number, tracks firmware versions and pushes over-the-air updates in staged batches. A stolen tracker can be revoked instantly.
Telemetry ingestion
Receives reports over MQTT, applies the validation chain described in Part 4 and writes every accepted fix before anything else happens.
Event processing
Geofence engine, rental state machine, anomaly model and ETA model each consume the stream and publish events such as “entered swim zone” or “overdue”.
Rules and alert router
Turns events into alerts with a severity, recipients and channels; handles acknowledgement, escalation and suppression of repeats.
API gateway
Authenticated REST and WebSocket API used by the staff app and dashboard, and available to the operator’s booking or point-of-sale system.
Storage
Time-series tables for fixes, spatial tables for zones and routes, relational tables for rentals, customers and staff, plus an append-only audit log.
Alerting and notifications
Channels are chosen by severity so that staff phones buzz for what matters and stay quiet otherwise.
| Channel | Used for | Target delivery |
|---|---|---|
| Dashboard | Every alert and event, with map focus | Under 5 seconds |
| App push | Warnings and criticals for staff on shift | Under 15 seconds |
| Criticals for supervisors and owner, with a live location link | Under 30 seconds | |
| SMS | Critical backup when data is unavailable | Under 60 seconds |
| Voice call | Unacknowledged after-hours theft alerts | After 2 minutes |
Dashboards
Live map
Every unit on the chart with state, rider, time left and trail. Alerts pin to the unit; click to follow it.
Rental timeline
Units down the side, hours across, each rental as a bar with dwell, alerts and return time marked.
Heatmap
Where riders actually go, by hour and day. Useful for redrawing zones and placing the patrol boat.
Reports
Utilisation per unit, overruns, alert counts, engine hours for servicing, exportable to PDF and Excel.
Storage and retention
Retention defaults follow data minimisation under PDPA 2010 and are configurable per operator.
| Data | Kept for |
|---|---|
| Raw position fixes | 90 days |
| Trip summaries and routes | 2 years |
| Customer contact details on rentals | 12 months, then anonymised |
| Incident evidence | Until the case is closed, plus 12 months |
| Audit log | 3 years |
| Backups | Daily, kept 30 days, restore tested monthly |
Security
Device identity
Unique X.509 certificate per tracker, stored in a secure element and rotated yearly.
Encryption
TLS 1.3 in transit for MQTT and HTTPS; AES-256 at rest for databases and backups.
Access control
Role-based access with two-factor sign-in for supervisors and admins.
Audit
Every view of rider data and every alert action is logged with who, when and from where.
Assurance
Independent penetration test before go-live, dependency scanning in every build.
Roles and permissions
| Capability | Dock staff | Supervisor | Operations manager | Owner / admin |
|---|---|---|---|---|
| See live map and units | Yes | Yes | Yes | Yes |
| Check out and check in | Yes | Yes | Yes | Yes |
| Acknowledge and resolve alerts | Warnings | All | All | All |
| See customer contact details | Active rentals only | Yes | Yes | Yes |
| Edit zones and rules | No | No | Yes | Yes |
| Reports and exports | No | Own shift | Yes | Yes |
| Manage users and devices | No | No | No | Yes |