Computer Vision Engineer

Role & responsibilities

• Design and assemble detection pipelines (detection, tracking, zone and line logic, OCR, alerting) for use cases like intrusion, PPE compliance, vehicle counting and license plate recognition.
• Build test harnesses that run pipelines against recorded footage and report precision, recall and false-alarm rates per camera, so changes are judged on numbers rather than demos.
• Collect, clean and label datasets, train and fine-tune YOLO-family models, and compare them against a held-out test set before anything ships.
• Profile throughput and memory of pipelines on target edge hardware.
• Tune pipelines on live customer cameras during rollout: thresholds, zones, sampling rates, lighting and night conditions.
• Debug field issues from logs and footage, write up root causes, and turn them into regression tests.

You should have

• 2 to 4 years building computer vision systems, with at least one model or pipeline you took from data to production (or a serious deployment-grade project).
• Strong Python, plus working knowledge of PyTorch, OpenCV and NumPy.
• Hands-on training and fine-tuning of object detection models (YOLO via Ultralytics or similar), including dataset preparation, augmentation and reading mAP, precision and recall honestly.
• Experience with multi-object tracking (ByteTrack, SORT or similar) and the logic built on top of it: counting, dwell time, line crossing.
• Familiarity with RTSP / IP camera streams and frame handling with OpenCV or FFmpeg.
• Daily use of AI coding agents, reviewing and owning every diff that gets merged.
• Gitflow.
• Comfort on Linux and the command line; basic Docker.

Desired Candidates

• Design and assemble detection pipelines (detection, tracking, zone and line logic, OCR, alerting) for use cases like intrusion, PPE compliance, vehicle counting and license plate recognition.
• Build test harnesses that run pipelines against recorded footage and report precision, recall and false-alarm rates per camera, so changes are judged on numbers rather than demos.
• Collect, clean and label datasets, train and fine-tune YOLO-family models, and compare them against a held-out test set before anything ships.
• Profile throughput and memory of pipelines on target edge hardware.
• Tune pipelines on live customer cameras during rollout: thresholds, zones, sampling rates, lighting and night conditions.
• Debug field issues from logs and footage, write up root causes, and turn them into regression tests.

You should have

• 2 to 4 years building computer vision systems, with at least one model or pipeline you took from data to production (or a serious deployment-grade project).
• Strong Python, plus working knowledge of PyTorch, OpenCV and NumPy.
• Hands-on training and fine-tuning of object detection models (YOLO via Ultralytics or similar), including dataset preparation, augmentation and reading mAP, precision and recall honestly.
• Experience with multi-object tracking (ByteTrack, SORT or similar) and the logic built on top of it: counting, dwell time, line crossing.
• Familiarity with RTSP / IP camera streams and frame handling with OpenCV or FFmpeg.
• Daily use of AI coding agents, reviewing and owning every diff that gets merged.
• Gitflow.
• Comfort on Linux and the command line; basic Docker.