A full-stack computer-vision studio

We're Apollume, a computer-vision and MLOps studio. We build and deploy production vision systems on GPU servers for heavy industry — safety monitoring, granulometry, and conveyor inspection — running on live camera feeds. We work across the full stack of a real CV system: training models, optimizing inference, building the backend, and shipping it so it runs every shift, not just in a notebook.

Our work spans the whole pipeline: training and fine-tuning detection, segmentation, and measurement models; optimizing them for real-time GPU inference (Triton, DeepStream, CUDA, TensorRT); and building the streaming layer, backend, and dashboards that turn a raw camera feed into an alert someone acts on. We care about the unglamorous parts — latency, calibration, drift, uptime — because that's what decides whether a vision system survives a real production line.

Led by Zakariae Mrabet — 42 / 1337 alumni · Arm Ambassador.