Lower costs and faster builds for a device lab

The challenge

The company depends on a fleet of devices collecting data. Every missed scan is lost revenue, so the fleet had to run reliably, update safely and collect clean data without costs running away.

What we did

• Built the R&D and lab operations function from scratch, with the device fleet split across two data centres for resilience.
• Automated installation, updates and baseline management, and published device documentation to the wiki automatically from the device metadata store.
• Fitted more devices into each rack, reducing how much rack space the business needed to grow.
• Refactored the video-collection pipeline to spread work more widely.
• Used Claude Code to rebuild the device container image, adding 64-bit support alongside 32-bit, and to fix a video encoding fault that had persisted for two to three years, working only on an isolated test device, never production.
• Set service-level objectives with first- and second-line on call and written post-incident reviews.

The result

• About £50,000 a year saved in data-centre costs over the first two years.
• Container image builds cut from about an hour to 25 minutes, with the image shrunk from over 1GB to under 500MB.
• The video-collection pipeline uses about a third less CPU under standard load.
• 98.7% of issues got a first response in time (target 90%) and 86.8% were resolved in time (target 80%).