We build AI vision systems that watch live industrial camera feeds and act on what they see — safety monitoring, real-time video pipelines, and automated inspection. Deployed on GPU servers, running on real cameras, every shift.
Real computer-vision systems running on live industrial camera feeds, built at an industrial AI R&D lab.
Detects PPE violations (helmet, hi-vis vest) and danger-zone intrusions on live camera feeds, with per-person tracking and automated alerts. Deployed on a GPU inference server against real RTSP streams.
Built and deployed at an industrial AI lab · client anonymized
Detects and measures rock from an overhead camera, reporting millimetre-accurate sizing via camera calibration — replacing manual spot-sampling on a processing line.
Built and deployed at an industrial AI lab · client anonymized
Continuous anomaly detection on moving conveyor belts, surfacing faults on a dashboard before they become downtime.
Built and deployed at an industrial AI lab · client anonymized
A streaming layer that ingests RTSP/HTTP camera feeds and runs interchangeable models, so new detection models deploy without rebuilding the pipeline.
Engagements are scoped, time-bounded, and delivered as running code.
PPE detection, restricted-zone alerts, and people/vehicle tracking on your existing CCTV/RTSP cameras.
From scoping to deployed system.
Multi-camera ingestion, GPU-optimized inference (Triton, DeepStream, CUDA), dashboards and alerting — shipped in Docker.
From scoping to deployed system.
Detection, measurement, and anomaly/inspection models trained on your data and deployed to run reliably in production.
From scoping to deployed system.
From a camera feed and a question to a system you can trust on every shift.
We look at your cameras, your environment, and the decision the system has to make. You get a straight answer on whether vision is the right tool — before any model is trained.
A working detector on your real footage, fast. We validate against the edge cases that actually happen on-site — bad light, occlusion, weather — not a clean test set.
GPU-optimized inference, packaged in Docker, ingesting your live RTSP feeds. Built to keep up with the cameras and to fail safely when one drops.
Dashboards and alerts so drift shows up before it becomes downtime. Models stay interchangeable, so a better one ships without rebuilding the pipeline.
The systems above were engineered at an industrial AI R&D lab — client anonymized. Teams and institutions we've built with and learned from:
Have a camera feed and a problem worth solving? Book a 20-minute call and we'll tell you straight whether vision is the right tool for it.
Led by Zakariae Mrabet — 42 / 1337 alumni · Arm Ambassador · production vision systems shipped at an industrial AI lab.