Book a call ↗

Case study / Security technology

Video intelligence.
Inside your perimeter.

An a16z-backed security-technology company that needed surveillance AI with no internet connection.

Book a 45‑minute discovery call
← All case studies

The difference it made

Value you can see.

Video AI that runs where the cameras are.

Response

Faster incident response.

Video AI runs locally, inside the customer site.

Releases

Routine releases.

Helm rollouts from an offline registry replaced releases that took days.

GPUs

Better use of every GPU.

Slicing and time-slicing share each card across streams.

The solution, at a glance

Offline AI.
Better use of every GPU.

The challengeWhat we changed
  • No internetAI inside the site
  • Idle GPUsShared GPU capacity
  • Slow releasesOffline installs & updates
Inside the customer siteNo internet required
  1. Video streamsLive feeds
  2. Stream routingMediaMTX
  3. AI inferenceShared GPUs
Offline platform
Local registry
Stream health GPU load
Engineering details

The challenge

  • Video AI had to run fully offline, inside the customer site
  • Expensive GPUs sat idle between streams
  • Releases into the air-gapped site were slow

The build

  • Air-gapped Kubernetes, installed and updated from an offline registry
  • NVIDIA GPU Operator with GPU slicing and time-slicing
  • MediaMTX scaled horizontally for live streams
  • Helm-standardised releases; DCGM telemetry in Prometheus and Grafana

Technology

KubernetesNVIDIA GPU OperatorGPU slicingMediaMTXHelmPrometheusOffline registry
The platform behind this caseProject 01 / Video surveillance
Continue exploringAI coaching that remembers.

Your next step

Have a similar
problem?

Forty-five minutes on your version of it. You leave with a proposed first step.

Book a 45‑minute discovery call