From proposal to live product in under five months

The result

Halo owns its control system and no longer has to rent one from a third party. It went live in September 2026, under five months after our first proposal in April.

Halo, the firms that hire its signals and their site crews can see every site, signal and fault in real time, from a phone. Remote control of sites is being added in stages, with the safety checks built in from the start.

The problem

Halo Traffic Systems is a UK company set up by highways and traffic management specialists. It makes portable traffic signals for roadworks, and each signal constantly reports its position, the light it is showing, battery health, signal strength and faults.

Halo wanted to own the system that turns all of that into something people can act on, and it had a launch date. An operator in a van at 2am needs to see a fault within seconds, and a wrong command at a live roadworks site is a safety issue.

What we did

We proposed a fixed-scope first release delivered in four weeks, then kept adding to it in phases. The system shows a live map of every deployed signal, a detail view for each one, and a list of what needs a crew. Halo staff, hire customers and contractor crews each see only what they should.

For the technical reader:

• Telemetry to screens. A Go service subscribes to Halo's MQTT broker, stores the telemetry and pushes every change to open screens in real time. New signals are detected automatically and held in a pool until assigned to a project.
• Scoped access. Phase 2 tightened ownership and the boundaries between organisations.
• Remote control with guard rails. Phase 3 adds remote operation in five controlled production milestones. It is off by default and must be enabled for both the organisation and the individual. There is a confirmation step before an all-red, and every action is audited and metered. Each milestone gets a full regression run and a security review before release.
• Built for fleet scale. Load testing showed per-site databases would fill a disk at fleet volume, so we redesigned storage around a time-series store, benchmarked for a 1,024-signal fleet with chart queries at 0.7 ms at p95.
• AI access. An authenticated MCP endpoint lets AI assistants read live fleet state through the same permission checks as the screens people use.

If you make equipment that works in the field and need to see and control it from anywhere, talk to us.